CVEN9407 Transport Demand Modelling
The Complete Study & Assessment Guide · T3 2026CVEN9407 Overview
- UNSW Sydney
- Term 3 2026
- Postgraduate
- 6 units of credit
- In person
What the course is about
CVEN9407 Transport Demand Modelling is a postgraduate course in the School of Civil and Environmental Engineering at UNSW Sydney, worth 6 units of credit and offered in Term 3 2026. On Moodle and in the lecture slides it appears under the shorter title Transport Modelling.
- Assessed by Five short Moodle quizzes worth 25%, a term project report worth 45% that is built in three parts, and a project presentation worth 30%.
- Hardest step Matching a model to the kind of outcome, then explaining what its coefficients mean for crashes in your own area.
- How to prepare Work each chapter's example by hand, then repeat the same method on your assigned crash records within the week.
- Pass condition Attending the Final Project presentation and achieving 50% in it is required to pass the course, in addition to a 50% overall mark.
How CVEN9407 is assessed
| Component | Weight | Format |
|---|---|---|
| Quizzes | 25% | Five Moodle quizzes across the term covering lecture and workshop content; each is 15 minutes long and taken at the start of the lecture, with cohort feedback given in class afterwards |
| Term Project Final Report | 45% | Individual project on an allocated dataset: a first deliverable on linear regression (up to 10 pages, 10 of 45 points) after the flexibility week; a second on count data and time series (up to 10 pages, 10 of 45 points) two weeks later; a final report adding discrete choice models (up to 40 pages, 25 of 45 points) in the last week |
| Final Project presentation · hurdle | 30% | Individual presentation in the last week of term: a 5-minute slot with 3 minutes to present and then questions. Attending and achieving 50% in the presentation is a hurdle to pass the course |
These three components and weights are the 2026 UNSW Handbook record for the course, and they total 100%. The assessment summary block on the Moodle course page lists a different split: quizzes 20%, draft report part 1 20%, draft report part 2 20% and a term project final report 40%, shown with start and due dates in 2025, so it appears to carry over from an earlier offering. If the two still disagree when you read this, ask the course convenor which split applies to you. Separately from the table, the Week 1 lecture slides set the pass mark for the course at 50% overall and a late penalty of 10% for each day past the due time. Attending the Final Project presentation and achieving 50% in it is a hurdle: without it you cannot pass the course, whatever your other marks.
Assessment structure
What CVEN9407 covers
Twelve chapters follow the course program from probability and estimation, through least squares and its diagnostics, to time series, count, logit and survival models, ending with a working manual for the crash data project reports and presentation.
Econometric Methodology and Transport Data
How a transport theory becomes an estimated and tested model, the step-by-step econometric method, observational data layouts and measurement scales (Week 1)The opening chapter sets out what econometrics adds to statistics: theory gives a direction, and estimation gives a magnitude with a measure of uncertainty.
- Statistics and econometrics as the two halves of the course
- The econometric method from theory to policy use
- Mathematical versus econometric models and the error term
- Observational data and its four layouts
- Ratio, interval, ordinal and nominal scales
- Regression, correlation and causation
Probability Distributions for Transport Data
Random variables and probability rules, centre and spread, skewness and kurtosis, the normal family, chi-square, t and F, and the Bernoulli, binomial and Poisson models for counts (Week 1)Every test and model later in the course rests on a handful of distributions, and this chapter assembles them.
- Sample spaces, events and the rules of probability
- Discrete and continuous random variables
- Mean, median, mode, variance and standard deviation
- Skewness, kurtosis and the Jarque-Bera test
- Normal areas and the central limit theorem
- Chi-square, t and F distributions
Estimators and Their Statistical Properties
Point and interval estimation, least squares, maximum likelihood and moments, bias and mean square error, BLUE, consistency and asymptotic efficiency (Week 2)An estimator is a rule applied to a sample, so it has a sampling distribution of its own, and this chapter judges estimators by that distribution.
- Estimators, estimates and sampling distributions
- Point and interval estimation
- Least squares, maximum likelihood and the method of moments
- Unbiasedness, bias and mean square error
- Best linear unbiased estimators
- Consistency and asymptotic properties
Hypothesis Testing, Error Types and Power
Null and alternative hypotheses, confidence interval and significance approaches, type I and type II errors, power, t tests on a mean, p values and two-sample comparisons (Weeks 1 to 2)Every coefficient in the project is eventually tested, and this chapter builds the testing logic on simple means first.
- Null and alternative hypotheses
- Confidence interval and test of significance approaches
- Type I error, type II error and power
- Confidence intervals for a mean and a variance
- One-sided and two-sided t tests
- P values and the 2-t rule
Two-Variable Regression and Ordinary Least Squares
Regression as conditional means, population and sample regression functions, the error term, least squares estimation, normal equations, residual properties and R squared (Week 2)This chapter fits the first regression line of the course by hand.
- Regression as a line of conditional means
- Population and sample regression functions
- The stochastic error term and the residual
- Least squares and the normal equations
- Slope and intercept formulas
- Total, explained and residual sums of squares
Inference and Prediction from a Fitted Regression
The normality assumption, maximum likelihood, standard errors, t tests and confidence intervals for the slope, the F ratio, mean and individual prediction, and reporting results (Weeks 2 to 3)A fitted line becomes evidence only when its uncertainty is measured.
- The classical normal linear regression model
- Estimating the error variance and standard errors
- Slope t tests and confidence intervals
- The F ratio and analysis of variance
- Mean and individual prediction
- Reporting a regression
Multiple Regression and Its Diagnostics
Partial effects, dummy variables and the dummy trap, adjusted R squared, multicollinearity and VIF, heteroscedasticity, autocorrelation and the limits of linear models for crash counts (Weeks 2 to 4)The first project deliverable is a multiple regression of crash frequency with full diagnostics, and this chapter supplies both halves. It explains partial coefficients, dummy coding of categories such as speed zone, and adjusted R squared.
- Partial regression coefficients
- Dummy variables and the dummy variable trap
- Adjusted R squared
- Multicollinearity and variance inflation factors
- Heteroscedasticity and the Breusch-Pagan test
- Autocorrelation and the Durbin-Watson statistic
Time Series Models for Monthly Crash Counts
Trend, seasonality and the autocorrelation function, stationarity and differencing, AR, MA and ARIMA models, forecasts and the time series section of the second report (Week 5)The time series step turns your area's crashes into a monthly or quarterly series and asks what its past says about its future.
- Trend, seasonality and memory
- The autocorrelation function
- Stationarity and differencing
- Autoregressive and moving average terms
- ARIMA order and model checking
Poisson and Negative Binomial Count Models
Why counts need their own model, Poisson regression with a log link, exposure offsets, overdispersion, the negative binomial variance, rate ratios and likelihood-based model choice (Week 3)Crash frequency is a count, and this chapter replaces the linear model with models built for counts.
- Why crash counts are not continuous outcomes
- Poisson regression and the log link
- Exposure offsets for segment length or traffic
- Overdispersion and its consequences
- The negative binomial model
- Incidence rate ratios
Discrete Choice and Logit Models
Random utility, choice sets and the multinomial logit, crash type as the outcome, independence of irrelevant alternatives, nested logit, ordered logit for severity and logit fit statistics (Weeks 7 to 9)The last econometric block models categorical outcomes, and the project applies it to crash type or crash severity.
- Random utility and the choice set
- The multinomial logit formula
- Base alternatives and alternative-specific constants
- Independence of irrelevant alternatives
- Nested logit
- Ordered logit for severity
Survival Analysis of Time Between Crashes
Gaps between consecutive crashes, survivor and hazard functions, exponential and Weibull models, the baseline hazard, Cox proportional hazards, hazard ratios and censoring (Weeks 9 to 10)Survival analysis asks how long until the next crash rather than how many crashes occur.
- Building gaps between consecutive crashes
- Survivor and hazard functions
- Exponential and Weibull models
- Baseline hazard and the Cox model
- Hazard ratios
- Censoring and the proportional hazards check
The Crash Data Project from Cleaning to Presentation
The Melbourne crash records and one assigned area, cleaning and descriptive analysis, the three written submissions, format and style rules, policy writing and the presentation hurdle (all weeks)The term project carries the written assessment and the presentation, so this chapter treats it as a working manual.
- The crash records and your assigned area
- Cleaning, recoding and fused information
- Descriptive analysis and data quality
- Contents of each report
- Format and style rules
- Writing policy implications
The Handbook describes it as a foundation in transport analysis and modelling, with a broad range of econometric techniques applied to transport demand and safety, and an emphasis on estimating models and using them for forecasting and interpretation.
The lecturer divides the teaching into two parts. The statistics part reviews probability, distributions, estimation and hypothesis testing.
The econometrics part begins with two-variable and multiple regression, works through heteroscedasticity and autocorrelation, and moves on to time series, count data, the logit family of discrete choice models and survival analysis.
Week 6 is the flexibility week with no teaching.
How the term project ties it together
The assessment is built around one individual project on real crash records from the Melbourne metropolitan area.
Each student is assigned a local government area and applies every method to that area's data: cleaning and describing it, fitting a linear regression with diagnostics, then time series and count models, then discrete choice and survival models, and finally drawing policy implications for road safety.
Two progress reports and a final report carry the written work, and a short presentation in the last week checks that the analysis is your own.
Attending the Final Project presentation and achieving 50% in it is a hurdle, so you cannot pass the course without it.
What this guide gives you
Twelve chapters follow the order of the course program, each with a figure, a worked example in invented numbers that you can recompute, and practice items with answers. The last chapter is a working manual for the project reports and the presentation.
Worked example · free
Is a crash count overdispersed? Poisson against negative binomial
- 1Compare variance with mean: 6.0 / 2.5 = 2.4, well above the value of 1 that a Poisson model requires, so the counts are overdispersed.
- 1Match the negative binomial variance, mean + alpha times mean squared, to the data: alpha = (6.0 - 2.5) / 2.5 squared = 3.5 / 6.25 = 0.56.
- 1Check it: 2.5 + 0.56 x 6.25 = 2.5 + 3.5 = 6.0, which reproduces the observed variance.
- 1Read the coefficient as a rate ratio: e to the power 0.30 = 1.35, so signalised intersections have about 35% more expected crashes; take its standard error from the negative binomial fit, because the Poisson one is too small.
Key terms
- Econometric model
- An econometric model is a mathematical relationship with a random error term added, so that its parameters can be estimated and tested from data.
- Ordinary least squares
- Ordinary least squares is the estimation method that picks the intercept and slopes giving the smallest possible sum of squared residuals.
- Residual
- A residual is the gap between an observed value and the value fitted for it by the sample regression line.
- Coefficient of determination
- The coefficient of determination, written R squared, is the share of total variation in the dependent variable that the regression explains.
- Type I error
- A type I error is the rejection of a null hypothesis that is in fact true, with probability equal to the significance level.
- Heteroscedasticity
- Heteroscedasticity is an error variance that changes across observations, leaving least squares unbiased but its usual standard errors wrong.
- Autocorrelation
- Autocorrelation is correlation between a series, or a model's errors, and their own earlier values.
- Variance inflation factor
- A variance inflation factor measures how much the variance of one coefficient is enlarged by that regressor's overlap with the others.
- Stationary series
- A stationary series is a time series whose mean, variance and autocorrelations stay the same over time.
- Overdispersion
- Overdispersion describes count data whose variance is larger than its mean, contradicting the Poisson model's assumption that the two are equal.
- Multinomial logit
- The multinomial logit is a model that gives each unordered outcome a probability equal to its exponentiated utility divided by the sum over all outcomes.
- Hazard ratio
- A hazard ratio is the factor by which a covariate multiplies the instantaneous rate of the next event in a proportional hazards model.
- Local government area
- A local government area is a council district; in the term project each student analyses the crash records of one assigned area.
CVEN9407 FAQ
How is the course assessed in 2026?
The UNSW Handbook lists three components. Quizzes are worth 25%: five Moodle quizzes, each 15 minutes at the start of a lecture. The term project final report is worth 45% and is earned in three parts worth 10, 10 and 25 of those points. The final project presentation is worth 30%. A different split on the Moodle page carries 2025 dates.
What exactly is the presentation hurdle?
In the last week of term each student gets a five-minute slot for the Final Project presentation: three minutes to present the project findings with slides, then questions. Attending and achieving 50% in it is a hurdle to pass the course. The Handbook says the task exists to make sure the report is not heavily AI generated, so expect questions about your own modelling choices.
How much statistics do I need before the course starts?
The first two weeks rebuild what the later models need: random variables, the normal, t, chi-square and F distributions, estimator properties, confidence intervals and hypothesis tests. Comfort with algebra and summation notation is enough to follow the least squares derivations.
Students who have not met hypothesis testing recently should rework those weeks carefully, because every later chapter tests coefficients the same way.
Which software do the lectures and project use?
The demonstration program runs regression and time series in R, logit models in Biogeme and the Cox model in R. The project brief supplies Google Colab notebooks with template code for each stage, and allows R, local Python, Biogeme, Stata or Excel instead, as long as your outputs match and your code can be handed to the lecturer on request.
What data does the term project use?
Individual crash records from the Melbourne metropolitan area, filtered to the one local government area you are assigned. Fields cover the crash itself, the people and vehicles involved, light, weather and surface conditions, and coordinates. You prepare separate files for regression, monthly counts, crash types as alternatives and the gaps between consecutive crashes.
Where do project reports usually lose marks?
No marking rubric is reproduced here, so take the brief's own priorities as the guide. It values interpretation over mechanics, final models presented with a justification, tables built cleanly with labels and significance indicators, PDF submissions, and critical comment on omitted factors and data quality. Raw software output, every trial model, and coefficients with no meaning attached all work against those priorities.
Is this a hard course?
It is a postgraduate course that covers a wide span: statistics review, linear regression with diagnostics, time series, count models, three logit models and survival analysis, all applied to a large real dataset in a single term. The presentation hurdle and the 50% overall pass mark both apply. The workload rewards keeping the project moving every week rather than cramming before the final report.
Questions students ask in each CVEN9407 chapter
- Why does the course start with methodology instead of formulas?
- Because every later model follows the same route: state a hypothesis, write it mathematically, add an error term, collect data, estimate, test, forecast and apply. Knowing which step you are on makes it clear why each table belongs in a report and what question it answers. Chapter 01 · Econometric Methodology and Transport Data
- Why does the sample variance divide by n minus one?
- Dividing by n would underestimate the population variance on average, because deviations are measured from the sample mean rather than the true mean. Dividing by n minus one removes that bias, which is why software reports it as the sample variance. Chapter 02 · Probability Distributions for Transport Data
- Can a biased estimator be the better choice?
- Yes. If its variance is much smaller, its mean square error can be lower than that of an unbiased rival. The lectures note that minimum mean square error becomes the criterion when the best unbiased estimator still has too much variance. Chapter 03 · Estimators and Their Statistical Properties
- What does it mean to fail to reject the null?
- Only that the sample gives no strong reason to reject it. The null is not proven, and another hypothesis might fit the data equally well. Report the estimate and its interval so a reader can see what values remain plausible. Chapter 04 · Hypothesis Testing, Error Types and Power
- Why are residuals squared rather than simply added?
- Positive and negative residuals would cancel out, so many different lines could make their sum zero. Squaring removes the sign and gives more weight to large misses, which produces a unique best-fitting line. Chapter 05 · Two-Variable Regression and Ordinary Least Squares
- Why is the error variance divided by n minus two?
- Two coefficients, the intercept and the slope, are estimated from the same data before the residuals are computed, which uses up two degrees of freedom. Dividing by n minus two makes the estimate of the error variance unbiased. Chapter 06 · Inference and Prediction from a Fitted Regression
- Does heteroscedasticity bias the regression coefficients?
- No. Least squares coefficients stay unbiased, but they are no longer the most precise estimators and the usual standard errors are wrong. That makes t tests and confidence intervals unreliable, which is why the brief asks for a heteroscedasticity test. Chapter 07 · Multiple Regression and Its Diagnostics
- Why must a series be stationary before fitting ARIMA terms?
- Autoregressive and moving average terms assume stable mean, variance and correlations. Fitted to a trending series, they can find spurious relationships created only by the shared trend. Differencing first, then modelling the stationary result, avoids that trap. Chapter 08 · Time Series Models for Monthly Crash Counts
- How do I know if my crash counts are overdispersed?
- Compare the variance of the counts with their mean, or the residual deviance with its degrees of freedom after fitting a Poisson model. Values well above 1 point to overdispersion, and a negative binomial fit with a clearly positive dispersion parameter confirms it. Chapter 09 · Poisson and Negative Binomial Count Models
- Why is one alternative set as the base?
- Only differences in utility affect logit probabilities, so adding the same constant to every utility changes nothing. Fixing one alternative's systematic utility at zero removes that redundancy, and every other constant and coefficient is then read relative to the base. Chapter 10 · Discrete Choice and Logit Models
- What does a hazard ratio greater than one mean for crashes?
- It means the condition raises the instantaneous chance of the next crash at every point in a gap. A ratio of 1.5 implies a 50% higher hazard, so gaps between crashes tend to be shorter while that condition holds. Chapter 11 · Survival Analysis of Time Between Crashes
- Can I use software other than the Colab notebooks?
- Yes. The brief names the Colab code as the official reference but allows R, local Python, Biogeme, Stata or Excel. You remain responsible for producing the same kind of output, and all code must be available to the lecturer on request. Chapter 12 · The Crash Data Project from Cleaning to Presentation
How to prepare for the assessments
Treat the course as one project with lectures attached. Each week, learn the method from the matching chapter, rework its example by hand so the formula is yours, and then run the same method on your own area's crash data before the next lecture. That rhythm keeps the reports small: the first deliverable is mostly written by the end of Week 5 if the regression and its diagnostics were done as they were taught.
Build a model log as you go, listing for each model the outcome, the data file, the specification and the one-sentence finding, because the final report must compare every approach. For the quizzes, revise the last two weeks' chapters and practise short calculations against a timer, since each quiz lasts 15 minutes.
For the presentation, rehearse explaining your choices aloud, and ask Sia to quiz you on why you picked a model or what a coefficient means; it will question your reasoning rather than write the analysis for you.
Your AI Engineering tutor for CVEN9407
Stuck on a hard CVEN9407 question? Sia is AskSia’s AI Engineering tutor — ask any CVEN9407 Transport Demand Modelling question and get a clear, step-by-step explanation grounded in how the course is actually taught and assessed. Read this whole study guide free, then take your hardest questions to Sia.
