Index: Synthetic Likelihood

Synthetic Likelihood

Routing Summary

Synthetic likelihood (Wood 2010, Nature) — a simulation-based / likelihood-free method for inference on noisy nonlinear dynamic models in the chaotic and near-chaotic regimes, where conventional likelihood collapses. Reduce data to phase-insensitive summary statistics, simulate to estimate their mean & covariance, and score fit with a multivariate-normal “synthetic likelihood” explored by MCMC. Contains 4 notes.

• New here / the big picture? → Synthetic Likelihood - Overview
• Why conventional likelihood fails; the Ricker map; choosing statistics? → Chaos and Phase-Insensitive Statistics
• The estimator $l_s(\theta )$, its properties, MCMC, MLE, diagnostics? → Synthetic Likelihood Construction
• Worked application (Nicholson’s blowflies, limit cycles)? → Nicholson’s Blowfly Application

Concept Map

Concept	Note	Type	Depends On	Key Result
Synthetic likelihood method; blowfly conclusion	Synthetic Likelihood - Overview	overview	Synthetic Likelihood Construction	Likelihood-free inference for chaotic dynamics via summary statistics
Likelihood collapse; phase-insensitive statistics; Ricker map	Chaos and Phase-Insensitive Statistics	concept	Synthetic Likelihood - Overview	$f_{\theta }(\mathbf{y},\mathbf{e})$ is intractable/irregular; judge fit on dynamic features, not local phase
MVN synthetic likelihood; estimation; MCMC; diagnostics	Synthetic Likelihood Construction	theorem	Chaos and Phase-Insensitive Statistics, MCMC Basics	$l_s(\theta )=-\frac{1}{2}(\mathbf{s}-{\hat{\mu }}_{\theta }{)}^{\top }{\hat{\Sigma }}_{\theta }^{-1}(\mathbf{s}-{\hat{\mu }}_{\theta })-\frac{1}{2}\log |{\hat{\Sigma }}_{\theta }|$
Gurney–Nisbet blowfly model; full vs demographic-only; stability	Nicholson’s Blowfly Application	example	Synthetic Likelihood Construction	Limit cycles perturbed by noise (ΔAIC > 1800 for full model)

Notes

• Synthetic Likelihood - Overview — CONTAINS: research question, one-line method, the four why-it-works properties, blowfly headline result, section/note map.
• Chaos and Phase-Insensitive Statistics — CONTAINS: scaled Ricker map (Eq. 1); likelihood-collapse argument (Fig. 1b-c); local-phase philosophy; design of summary statistics (autocovariances, AR/regression coefficients).
• Synthetic Likelihood Construction — CONTAINS: MVN approximation (Eq. 2); 4-step evaluation algorithm + $l_s(\theta )$ formula; properties (smoothness, invariance, robustness, $N_r\to \infty$); Metropolis–Hastings MCMC; MLE via quadratic regression; AIC/GLRT; ${\chi }_{\dim (\mathbf{s})}^2$ checking diagnostic.
• Nicholson’s Blowfly Application — CONTAINS: Gurney–Nisbet model (Eqs. 3-4) + stochastic discretization; blowfly summary statistics; full-vs-demographic ${\chi }^2$/AIC comparison (Fig. 3); stability-diagram analysis (Fig. 4) → intrinsic limit cycles.

Sources

• Wood 2010 - Statistical Inference for Noisy Nonlinear Ecological Dynamic Systems — Wood, S.N. (2010), Statistical inference for noisy nonlinear ecological dynamic systems, Nature 466(7310):1102–1104. DOI:10.1038/nature09319.

See Also

• Approximate Bayesian Computation for ABMs — the closest likelihood-free relative (acceptance-threshold vs. MVN likelihood)
• Method of Simulated Moments / Indirect Inference / Simulated Moments Estimation - Overview — summary-statistic / moment matching by simulation
• MCMC Basics — the sampler used to explore $l_s$
• Model Comparison — AIC / likelihood-ratio model selection