ECON0006 Chap.10 Synthetic Data and Networks
Synthetic Data and Networks
Synthetic Data and Networks examines controlled random generation and interacting structures. A synthetic dataset needs a declared distribution, dependence pattern, parameters and replication plan. A network needs node and edge meanings, including direction and weight conventions. Statistics from one realised graph describe that graph; repeated simulations estimate properties of the generating model.
Reproducibility connects code exploration to written mathematical reasoning.
What this chapter covers
- 01
Data-generating processes
- 02
Random variation
- 03
Replication
- 04
Node semantics
- 05
Directed and undirected edges
- 06
Degree counts
- 07
Model versus realisation
Worked example · free
Audit a small undirected network
- 2Add the degrees to obtain 10 endpoints.
- 1Divide by two because each undirected link has two endpoints.
- 1Report five links and the identity used.
Key terms
- Synthetic data
- Data generated from a specified mathematical or computational process.
- Network node
- The entity represented as a vertex in a network.
- Network edge
- A defined relationship connecting two nodes, possibly with direction or weight.
Synthetic Data and Networks FAQ
Why repeat a random simulation?
One realisation contains sampling variation. Repetition reveals the distribution of a statistic and separates a recurring model pattern from an unusual draw. Report both the repeated pattern and its uncertainty.
When should a network be directed?
Use direction when the relationship has an origin and destination, such as exports or influence. State the convention before calculating incoming or outgoing degree so an arrow's meaning remains unambiguous.
Does a highly connected node prove influence?
No. Degree is a local structural count. Influence depends on the process operating over links, direction, weights and the question being asked. Centrality measures answer different structural questions and need interpretation.
What is a useful hand check before a large simulation?
Run the procedure on a tiny dataset or network whose draws, links and statistic can all be listed manually. Compare the hand result with the implemented output before increasing sample size or repetitions. Agreement separates ordinary random variation from a coding or definition error, while a mismatch can be traced without the noise and scale of a full experiment.
Exam move
Write the data-generating process before interpreting a synthetic result. Record distributions, dependence, parameters and the replication or seed convention. For networks, define one node and one edge in the economic setting, then mark direction and weights. Hand-count a small graph and reconcile degree totals with link totals before using software.
Distinguish an exact fact about one graph from an average across simulations and from a causal claim about observed data. Review mistakes by undocumented randomness, marginals treated as a joint model, directed links counted as undirected and one unusual realisation presented as a universal pattern. Treat a synthetic dataset as the output of a declared model, not as a collection of convenient numbers.
Write distributions, parameters, sample size and dependence before generating anything. State which features remain fixed across replications and which are redrawn. If variables share a shock or have correlated components, encode that relationship rather than sampling correct marginals independently. Choose a statistic linked to the question and predict how the generating rule should affect it.
This prediction creates a benchmark for detecting implementation errors or an unusual realisation. Use a replication table with one row per simulated dataset or network. Store the chosen statistic, then summarise its centre, variation and extreme outcomes across rows. One run can illustrate a mechanism but cannot establish the distribution of the result.
Record a seed convention when reproducibility matters and vary sample size separately from random repetition. If results change sharply with sample size, explain whether the change reflects reduced sampling variation, a finite-size network property or a different model input. Keep those interpretations distinct. For networks, define node, edge, direction and weight before calculating structure.
On a small graph, list every edge and count incident connections beside each node. For an undirected simple graph, compare the degree total with twice the edge count. For directed graphs, separate incoming and outgoing degree; for weighted graphs, distinguish connection count from strength. State how self-links and repeated edges are treated.
Only then interpret centrality or connectivity, since different statistics measure immediate neighbours, path position or another structural feature and need not rank nodes identically. Finish each exercise by identifying the level of the claim. An exact degree count describes the realised graph. An average across replications estimates a quantity under the generating model.
A comparison with observed economic data raises additional measurement and causal questions that simulation alone does not answer. Write pseudo-code in the order generate, calculate, store, repeat and summarise, with parameter choices outside the loop unless they are deliberately varied.
This transparent sequence remains useful in a closed-book response even when software is unavailable and makes every assumption open to audit.
Working through Synthetic Data and Networks in ECON0006? Sia is AskSia’s AI Economics tutor — ask any ECON0006 Synthetic Data and Networks question and get a clear, step-by-step explanation grounded in how ECON0006 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.