Forecasting traffic is
widely regarded to be the first step in the process of any planning or
renovation model. It is not only the primary step in determining the
practicality of a project by figuring out the scale of benefit rendered by the
outcome of a project, but also the extent of its criticality. Travel demand is
also an important socio-economic indicator of development, functional capacity
and administrative quality within an executive region. As with forecasting of
any and every form, travel demand or traffic forecasting is a tricky business.
The complexity of estimating this quantity is rendered no less intricate due to
its multifaceted nature: travel demand is a function of a number of factors.
These factors not only vary on the basis of location and the time period of the
forecast, but also across the intrinsic values of the predictors used in the
model.
Accuracy often is not the
most sought after skill among forecasters, says Silver. The consequence of this
becomes particularly dangerous in time-sensitive events like hurricane Katrina.
Residents of New Orleans, a laid back Louisianan city, like people elsewhere,
took neither the hurricane prediction nor the word of the city mayor very
seriously, which led to the vastly infamous and large-scale devastation.
Disbelief over superficially unimportant issues like weather, when aggregated
over time, leads to skepticism about something as major as an all-engulfing
hurricane. This, in today’s day and age, is a very challenging problem to
resolve. This is due to the culmination of years of, for lack of a better word,
sloppy priorities for evaluating forecasts. Forecasters belonging to most
walks, are judged harsher over their skills of presentation and precision than
accuracy and honesty. Though precision ostensibly looks like a highly
sophisticated metric for the quality of a forecast, it often is bought at the
price of accuracy. Precision should be used as a tool to adjudge the accuracy
and confidence of a forecast, not the result of the forecast itself. Rampant
misuse of precision and over-confidence in one’s predictions leads to the mass
delusion of futility of the very exercise. For example, if weather forecasters
are made to compete with climatologists over defending the precision of their
estimate, objectively, they would lose more often than not. This wouldn’t make
for a very convincing (read profitable) weather forecast show. It thus paves
way for a discomfiting circular logic. Weathermen are not looking to hire
overtly accurate weather forecasters because people have relatively lower
confidence in TV weather forecasts anyway; people have little confidence in TV
weather forecasts because they are not accurate most of the time. Similarly,
traffic forecasts are used more as a tool for validating the personal benefits
of pre-decided policies, rather than as a tool to objectively influence the
decision-making process. Also, if the estimates made by forecasting pundits are
presented in the form of a range than exact numbers, it would imply that policy
makers (who, in this case, are evacuators and other disaster management
personnel) have to formulate solutions and prepare for more than one scenario.
In some contexts, they might even be expected to brace for more than a finite
number of setups. In a world used to drawing distinct borderlines between
abstract events, this becomes an uncomfortable predicament. The cost of
addressing this hardship was paid by those trapped in the hurricane.
In
the chapter, “A Catastrophic Failure of Prediction”, Nate Silver outlines the
series of events which led to the disreputable economic crisis of 2008.
Housing, which since the very advent of the science of analyzing macroeconomic
influence of different market commodities, was never chalked up to be a
particularly lucrative investment. This directly transcribes its decidedly safe
nature. The question which should have been asked, but no one did, was “Really
how safe is safe?”. Credit rating agencies exist to answer precisely this
question. Their sole purpose is to quantify the riskiness of investments (and
mortgages). The process of answering this question is clearly not as
straightforward as one, or in hindsight, the rating agencies might have hoped.
The very intricate route to reach a conclusive answer to this question was
conveniently ignored by the risk rating agencies as a result of the inherently
“safe” nature of housing investments. Proprietors of this type of investments
are generally considered law-abiding, high income citizens, and their
credit-worthiness was believed to be solely a factor of outrageously under-researched
and over-simplified ingredients like geographical location and income (for
example, a software scion in Silicon Valley versus a damp shack dweller in
Arkansas). Preposterously large and protracted ranges of credit-worthiness
scores were abridged into rating ranges as a result of this faux pas. That
neither investors, nor rating agencies questioned the quantitative nature of
the bond ratings is egregious and that it happened because even a confidence
score of 99% has a 1% chance of failing is a very flimsy line of defense
indeed. It has been conclusively proven through numeric simulations,
distributional graphics, and highly sophisticated multi-dimensional
computations, that given a sufficiently large dataset, a 1% risk is indeed
transformed into a 100 in 10000 statistic. If historical data is used to
diagnose the validity of a statistical model, the given rate of risk is inbuilt.
That is to say that if the forecaster has a history of successful predictions
over a statistically significant period of time as well as an adequately large
dataset, the risk percent is a known and quotable statistic. It should be noted
that underestimation of risk in such magnanimously high stake markets is seldom
a result of sampling biases and errors. It is instead, as Dean Baker said,
‘baked into the system’. It is usually a result of a faulty statistical model fueled
with defective and incorrectly hypothesized assumptions. This gives the entire
spectacle a wildly criminal undertone. The inviolable nature of this mortgage
market did not invite a lot of investigative probing, which led to its slow but
exponential rise in attracting colossal amounts of investments, hedge funds,
speculation bets and the likes of it. Its share of fixed assets in the economy
rose so much so that it became one of the major drivers of the economy.
The algorithmic structure of
assigning the ratings to pools of mortgages to “bet” on reveals the expected percent
default within each pool of investments. Assumptions are then made as to which
pool of risk each combination of mortgages belong to. What the rating agencies
failed to take into consideration was a very obviously delineable and existing
connection between seemingly unrelated housing mortgages (and consequently
their default rates): the skyrocketing prices of houses with no significant
rise in incomes, essentially, what we call a Bubble. The money spent on buying
these houses, instead of having a market currency value, was reduced to mere
numbers further fueled by the constantly increasing number of nonpayers. Another
argument for an apparently unaccounted for noise (error) hauntingly converting
the 1% risk into a glaring reality could be the possible absence of
context-specific evidence which could not have alerted any active seekers to
address the inherent risk. This is not unlike how a driver with an untarnished,
three-decade old driving record is unequipped with any evidence to adjudge how
safely he might drive drunk after a party one night. But this situation was not
applicable in case of the housing bubble fiasco. This is because despite having
a precedence of two other similar housing bubbles almost 50 years ago, the
first in USA, and another eerily similar case in Japan, the rise of this
flourishing market continued without any checks. Partly to blame was the investors’
proclivity to the optimistic aversion from inspecting the number of defaulters
within this market which prevented private rating agencies from performing
basic legal quality assurance requisites. Most investment combinations that
used to be remotely composed of housing mortgages and/or related to the housing
sector was automatically given the highest (signifying the safest) rating
(AAA). Any attempt at investigating the underlying cause of the increasingly
prevalent AAA bonds in the market was either disregarded, discouraged or simply
quelled with an air of poised incredulity.
After having established the
necessity of paying ample attention to the often unnoticed but
context-sensitive factors lurking in a dataset, one should also consider
dividing the data into the appropriate resolution to make the forecast
practical and avoid over-generalization of the derived results. Silver
addresses this in the chapter “For Years You’ve Been Telling Us That Rain is
Green”. The failure to first conclusively envisage the hurricane Katrina, and
then effectively communicating the necessity of evacuating New Orleans left
thousands dead and many more devastated. Factors like consumer/client interest
in the interpretability of the forecast, impact of the decision relating to the
forecast, competition in a race to throw out the most popular and easily
inferred forecast, all affect the methods and effort which go into forecasting.
Recognizing that pooling a lot of data together would only gradually give way
to chaos taking over is crucial to avoid overly generalized solutions to
complex research questions. In many cases, this might even lead to incorrectly
classifying data into categories which are unsuitable given their number of
dimensions. Transportation systems and factors leading to any major change in
the demand for traffic facilities are symbolized by utility functions which are
dynamic in nature, ever changing and sensitive to changes. These changes might
not be completely intuitive, and can evade from notice now, more than ever,
given the vast amount of data modern day researchers have access to. Chaos
theory also points out the non-linear nature of these systems meaning that they
might be exponentially affecting some component of the utility function.
Advanced computational facilities these days enable forecasters to speed up
multi-dimensional calculations, a skill which is only useful if applied in a
carefully designed, context-specific manner.
Another factor of importance
is translation of data. In the field of forecasting, or in this case, in the
field of traffic demand forecasting, translation of data and results sourced
from a different point in space and time plays a pivotal role in affecting the
final inference derived from the data base. Silver, in the chapter “All I Care
About is W’s and L’s” of his book, mentions Baseball
Prospectus which enumerates statistics related to baseball players, both
major and minor league. Numbers corresponding to the minor league players are
skillfully standardized to make them comparable to those corresponding to the
major league players. Results often sourced from analyses, studies and
predictions carried out in conditions unrelated to the context within
examination created a bias in our method of forecast. On one hand, it is
important to keep track of the advancements in forecasting methodologies
elsewhere and incorporating these approaches with suitable caution to our own
circumstances, it is also essential to learn to segregate such results and
avoid them from introducing unnecessary noise and prejudice in our own
calculations. Some events are plain random; a consequence of belonging to a
larger, more untamed set of disconnected data, and are capable of causing
significant kinks in our prediction results. A number of factors differentiate
major and minor league players from each other, for example, size of field, frequency
of games, etc. It is therefore imperative that wins and losses from each
category of the game be normalized to make an “equal-ground” comparison between
them. In a game like baseball, which is often played in fields of non-standard
dimensions, it is essential that such calibrations be made not only between
major and minor leagues, but often even between games belonging to the same
league categories. How this extends to forecasts related to transportation and
traffic demands is that methods and forecasts which have been historically
proven to be correct may not necessarily work in a different setting, a
different time period, or even a different demographic group. Baseball
Prospectus, for instance, used ‘Park Scores’ to evaluate and homogenize scores sourced
from games played in each park. Similar empirical studies need to be carried
out when calculating demand within a certain environment composed of different
temporal, spatial and heuristic elements like social groups inhabiting an area,
recognizing the key demographic which is expected to use a certain
transportation facility, number of years the forecast is being calculated for,
mode for which the transport demand model is being calculated for, etc. These
scores should then be factored in the analysis with appropriate consideration
when forecasting demand for the respective conditions.
Traffic forecasts
essentially play the role of rating agencies in a world composed of federal and
state departments of transportation, private construction contractors,
political policy makers, urban planners, etc. To account for the numerous influential
nuances embedded in the socioeconomic and temporal data driving transportation
forecast models, it is vital that context-sensitive methodologies be adopted.
If forecasting ethics are isolated from political uses of transportation model
forecast results, arriving at objective predictions based solely on available
data, particularly in developed countries like the USA seems to be a viable and
practical endeavor. However, the presence and interests of major stakeholders
govern such forecasts. Analysts, in such cases, are expected to devote more
time in coming up with technically sound defense arguments justifying a set of
predetermined outcomes than ensuring the accuracy of the forecast itself.
Resource allocation to various transportation projects benefits a number of
sections of the society. Although there may be grounds of rationalizing such
benefits from a political perspective, the purpose of forecasting traffic
demand should not, in strict terms, be regulated by such extenuating
validations. On occasions, forecast specialists employed in firms tweak certain
assumptions to which the results of the forecast inherently adhere to for
fulfilling self-serving purposes: for instance, in order to rake in subsequent
contracts, the existence of which is tied to the necessity of the initial step
being projected indispensable. As forecasts are, by definition, unverifiable if
the underlying assumptions are somehow warranted, they are basically used as a
primary obligatory step instead of a decision which fundamentally drives the
decision making procedure [1]. The rampant abuse of estimation methodologies
has made people cynical towards the very necessity of this procedure. As a
result, policy makers may soon advocate the removal of this step in its
entirety instead of making the process transparent and accessible to the
public. This would crown into the blatant abuse of political mandate and
stakeholders will solely be responsible for authenticating the requirement of
any future public project. Thus, resolutions to address this are need of the
hour. As suggested by Wach (1990), aware and educated masses well-equipped to
question the authenticity and accuracy of the forecast results will provide a
greater impetus to forecasting agencies as well as political and
entrepreneurial stakeholders to function more honorably. Also, conceiving and
acknowledging the ambiguity in the forecasters’ code of ethics and professional
practices would go a long way in addressing the dilemma forecasting personnel
experience in the face of being honor bound to serve their employer as well as
being true to professional integrity. Hartgen (2013) explores the techniques
European and Australian forecasters use to address the uncertainties in the
values of the variables shaping their transportation model predictions. Uncertainty
logs are developed to quantify the indecisions associated with the values of
each randomly distributed variable. These quantitative measures are then
classified into ranks (like the rating agencies did with the risk associated
with each bond). A list of decisions linked with each of these ranks serves as
an advisory tool to make decisions about the practicality and necessity of the
project being undertaken. The risk-based decision module follows a rubric which
lists various recommendations so as to facilitate the decision making process,
for example, if a specific project could be undertaken by tweaking the
policies, or changing the size and scale of construction, etc. Hartgen points
out the necessity of calling out the ethical concerns related to forecasting
practices instead of fixating solely upon modifying the underlying structure of
the forecasting methodologies from modal to topical. Like Silver, he also
suggests presenting the result of travel demand models being presented in a probabilistic
form and/or ranges of possible scenarios rather than mere numbers. He notes
that this will increase the quality of these forecasts in terms of accuracy, as
well as preparedness on the part of the contractors. He has formulated a rubric
for assisting the process of identifying inaccuracies in the forecast results
which can be used by almost every hierarchy of the stakeholders- public,
journalists, political beneficiaries, contractors, engineers as well as
analysts. Unrealistic and unverified assumptions often are the most serious
perpetrators of spitting out a glaringly flawed forecast. The validity of these
assumptions should not be based purely on their justifiability. In an uncertain
socioeconomic paradigm, which exists for any forecast made for a time period 20
years into the future, any assumption can be ascertained citing the vast cloud
of improbability. These assumptions simply serve as a computational convenience
and should be treated as such.
Using untested methodological
advancements in the field of travel demand modelling forecasts, without
carrying out sufficient reliability tests exclusive to the context of the
forecast should be discouraged. Care should be taken to make sufficient
corrections for temporal bias and sampling errors in the traffic behavior data
when using the four step model as the data and assumptions related to travel behavior
is obtained from different spatial points across an area, but not necessarily
from different points in time. Overfitting the data, as both Hartgen and Silver
pointed out, is another major but tremendously commonplace misapplication of
statistical models. Overfitting creates early breeding grounds for insinuating
bogus relationships within variables in a database. One might think that such spurious
relationships could be easily spotted and expelled from the model results but
in reality, it is not as straightforward as that. Real world data is much more
amorphous and noisy to enable spotting a visibly conspicuous relationship. In fact,
some of these spurious correlations may in fact not be intuitively as apocryphal
at all. Silver quotes the winner of the Super Bowl being considered a major
predictor for the development of the economy for the better part of 1990’s. The
hypothesis had excellent R-squared and P-values. The model even performed well
in “predicting” the GDP growth for a few years before starting to fail and
being called out for its co-incidental and correlation-without-causation
nature. Theoretically, the probability of the relationship being merely due to
chance was less than 1 in 4700000. What is interesting is that these figures
could as easily be generated by fitting a model of chicken production in Uganda
and the economy of USA (this effect is more commonly known as the Butterfly
effect). Similar is the effect of personal biases introduced during sampling or
modelling by the analyst producing intuitionally sound forecasts. Utmost care
should therefore be taken to ensure the comprehensiveness of an analysis, and
explicit post-hoc diagnostics should be encouraged to minimize the risk of
ending up with a contrived forecast. This again fortifies the necessity of
reporting results accompanied by their respective margins of error.
Zhang et al. (2012) in their
study about peak temporal traffic trend forecasting, compared different
non-parametric models to arrive at the most effective and least computationally
intensive method to provide real time peak hour traffic forecasts using
historic peak time traffic data [2]. It should be noted that even for models
employing non parametric methods (in this case, Least Square Support Vector
Machines) to analyze large time series data sets, there is considerable noise
on days displaying more haphazard peak hour traffic (Thursdays, for instance).
Historic data too fails to predict with significant accuracy the real time
traffic data to be expected during these hours. This is reminiscent of the
erratic weather forecasts made 10 days earlier to the target date that Silver
mentioned in the chapter “All I Care About Is Wins and Losses”. Forecasts made
ten days prior to the target date and promptly rolled out through savvy
interfaces were based mostly on moving averages of historic weather data.
Hardly any refined analysis actually went into the calculations leading to
these forecasts. The forecasters hardly have any faith in these numbers. More
often than not, these numbers fail to resemble the more cogently forecasted
predictions (based on climatic and temporal data, often within a week of the
target date) but perchance do seem intuitive to the actual weather encountered
on the target dates.
Building on that premise, in
a race to make forecasts appear more precise and larger than life than they
really are, individuals or agencies often try to distract an observer away from
the number of times the prediction failed. The fail percentage, or ‘risk’,
associated with a forecast is as important as the accuracy, result and the
confidence level of the forecast itself. Selective reporting of results is
fundamentally unethical. But when the stakes are not too high, for example a TV
show prophesying election results, or when a forecaster is instructed to mold
his/her predictions to fit a specific frame, failed predictions are hardly ever
discussed. It should be noted that a lot of such discussions, like the TV panel
anticipating election results in weeks leading up to the final election day are
substantially a form of feedback data as opposed to the mere critique analysis
they usually guise themselves in. Many of these ‘analyses’ have stark
underlying flaws in them, like using small sample sizes, evidence opposing
their hypotheses conveniently being overlooked, etc. While some of these
predictions have to be right only once for their predictor to be regarded as a
highly gifted political analyst, many of these experts have made multiple,
often blatantly contrasting, claims based on incidental evidences. But this
elitist stance may not work entirely in the favor of political scientists
either. What it depends upon is whether their forecasting approach is based on
pools of contingent data or do they follow a one size fits all methodology to
arrive at their conclusion. Sometimes, information obtained from diverse and
unrelated strands are woven together by keen observers to predict certain
outcomes. A trite example for this supposition would be the failure of many a
political scientist to predict the collapse of USSR. The dissimilar pools of
data (in this case, news) were not even contrary to each other that they would
have instigated two opposing schools of political prediction pedagogy. Instead,
people who predicted the demise of the union merely fortified their conclusion
by assimilating the data gathered from these multiple sources. As mentioned in
the chapter “Are You Smarter than a TV Pundit” (a pun on the popular TV show,
“Are You Smarter than a Fifth Grader”, cleverly sneering the likes of McLaughlin)
by Silver, “fox like” forecasters, who are scrappy about locating information and
interlacing them together to form a comprehensive story usually have a higher
success rate when it comes to prediction. In spite of this, their predictions
hardly make the headlines, probably due to the absence of an overbearing
cockiness about their predictions. Also, their predictions are typically based
on unintuitive bits of information strewn together, often culminating into
inexplicable and far too complicated empirical derivations. An average reader
or viewer does not have the patience nor acumen to try to sift through the
humdrum of proofs therein. These people also do not characteristically make
particularly charming TV show guests. The roster of clues they have to offer
incorporate ideas from multidisciplinary sources, their statistical models are
highly sensitive to new pieces of information and due to these reasons, they
are far too cautious about their predictions. Their predictions also frequently
fail to resolve a lot of seemingly
related surrogate questions mostly due to the reason that these questions are
just that to them: unresolvable, given
the current set of data.
Another category of
forecasters, that is, “Hedgehogs” to quote the famous UC Berkeley psychiatrist
Dr. Tetlock, are what you and I would call the “Alpha” forecasters. They are
dedicated analysts, often trained to offer predictions based on tested
theories. Additionally, they are mostly career forecasters who deal with
limited areas within forecasting (in extension, this type of personality is
often found frequenting within professional urban planners, who are exclusively
trained to function within specific fiduciaries). People with such highly
devoted expertise often belittle outsiders’ opinions, ultimately rejecting a
probable pool of additional information a keen, scrappy fox might have gathered
and to offer. In such cases, traces of new data is not viewed as a potential
resource to change the pre-existing theories or statistical model, but to refine
the current (actually, age old and time tested) model. This often leads them to
ignoring crucial changes of the current time and day which might lead to what
Silver earlier referred to as a catastrophical failure of prediction.
Acknowledging the chaos within the predicting variables is the first step
towards attempting to evaluate it and “hedgehogs” are very reluctant to embrace
the anarchy within their analytical turf. The exceedingly imposing inferences
made by “hedgehogs” in their predictions make for excellent TV guests; they
exude the sort of confidence which accompanies precise forecasts, irrespective
of accuracy.
Instinctively, the secret to
develop oneself into a better forecaster clearly lies in being ‘foxy’. What
this means, implores Silver candidly, is to assess probabilities instead of
figures. This means, that in many cases, the evaluator might be left with a
range of outcomes as wide as almost half of all possible results. A ‘hedgehog’
might argue that this is a pre-conceived excuse to corroborate the failure of
one’s predictions. What they will most likely fail to take into account is that
this range will prevent a forecaster from quoting a figure as absurdly off the
mark as predicting that Republicans would win 100 seats within the Electoral
College in the 2010 presidential elections when it won 63. This is because
individual P-values pertaining to singular value for the dependent variable are
often misleading when a statistical model is not corrected for selectivity
bias, random variable biases and unidentified panels in the data. On the other
hand, the likelihood of a range of values within a reasonably defined
confidence interval being as preposterously faulty as the one mentioned above
is meagre (in which case, there either exists a capital flaw in the methodology
of the model forecast, or the number of unidentified lurking variables is too
large to be accounted for). This is because the combined likelihood of a range
of values, even after being conservatively corrected for multiple comparisons
and post-hoc analysis, will hardly administer a misleading prediction, even if
it is not aggressively precise. Although, the results from this foxy mechanism
of forecasting are useful and even applicable to real life problems spread over
a longer period of time and a sufficiently large dataset, it might be
impractical and, to some extent, cumbersome to substantiate the results spewed
by a range of possible outcomes. When lives are not at stake, a more
conservative or ‘hedgehog’ way of working might be vouched for given ample
empirical evidence to prove its veracity. But as extensively as practical, the
consumer(s) should espouse their faith in the more soundly principled
probabilistic prediction to avoid making high stake blunders in attenuating
circumstances like threat to life (field of medicine) or the economy.
Khaled et al. (2003)
addressed the issues surrounding lack of credible data sources in developing
countries and its impact on traffic forecasting, and consequently on the lack
of resources to conduct cost benefit studies corresponding to numerous land-use
and urban planning projects [3]. This, coupled with the high urban population
densities, underdeveloped roadway facilities for non-motorized vehicles, high
crash and traffic fatality rates, degrading environmental conditions and
saturated conditions of traffic operations in developing countries, expounds
the scope of financial losses if an unplanned and poorly analyzed
transportation project is undertaken. Moreover, with continually rising incomes
the need of a thoroughly planned infrastructure and traffic management system
is becoming the need of the hour. The four step method of urban planning,
essentially developed for first world countries is ineffective and trite way to
address the entirely different heuristics of developing nations. The trip generation
method of the four step model requires socioeconomic data which predicts the
demand for transportation facilities while there is no consideration for
quantitative metrics like travel time and roadway capacity which are subject to
highly distinct and discernible differences in developing countries depending
upon a number of traffic interruptive influences. Also, the population and land
use patterns change more rapidly in a developing economy than that in developed
countries, making predictions well into the future (usually necessities and
magnitude regarding transportation facilities are made twenty years into the
future) subject to a variety of unknowns. To counter this, Khaled et al.
suggested modelling the respective urban network into the TransCAD software,
superimposing demographic and land use data from a GIS shapefile, and creating
skim matrices of the origin-destination pairs thus generated. The utility
functions to be used in distributing the trips and factoring the mode splits
would be generated by an empirical expression, the summation of the individual
products of trips T and a proportion P determined by the stochastic nature of
the destination. The stochastic nature of the proportion variable is due to estimating
traffic from new generators, varying land use patterns and changing demographic
statistics of the region in question.
While Khaled et al. investigated
the experiential relationships between transportation-demand modelling in
developing countries; Naess et al. (2014) examined sources and causes of
forecasting inaccuracies in transportation modelling in Scandinavian countries
[4]. They have analyzed and quantified errors arising from a number of methodical
errors while statistically calculating future travel demand models. Survey data
was used to explore extant practices in the field of transportation
forecasting. Insights were sought within this data to identify underlying
sources through which common inaccuracies like optimism bias, strategic misrepresentation,
sampling errors, etc. could creep into the model. In order to avoid
generalizing the results obtained from this study, it should be noted that the
region addressed in this study, namely Scandinavia, is a developed economy. It exists
at its peak of socioeconomic prosperity and possesses highly advanced systems
of traffic system and transportation facilities. Additionally, the traffic
forecasting agencies are institutionalized to most extent, meaning that there
lies certain uniformity between how these forecasting agencies within the same
department operate. Overestimation of future traffic demand was found to be a
major issue in most European countries. This is widely referred to as Optimism
Bias. Optimism bias also includes the underestimation of construction costs. This
combination of issues is a common occurrence, which was also scrutinized by
Hartgen and Wach in their papers critiquing prevailing transportation
forecasting techniques and ethics [5]. Other insights which could be drawn from
the Scandinavian questionnaires were that most of their forecasting agencies
almost unanimously agreed that ontological explanations like delays in
construction, unpredictable land use development and development of some unforeseen
transportation infrastructure lurked behind failed predictions of traffic
demand models. That unexpected and unpredictable geopolitical trajectories and
vastly different vested interests of political and business groups were
responsible for the uncertainties in the predictive models and significantly
accurate predictions could not be made for demand models as far as 10 years
into the future based on the data we have access to today. Interestingly, there
is hardly any mention of probabilistic values and/or ranges of predicted values
for the demand models which might prove useful in explaining their undisputed
distrust in predictive models.