15 Final remakrs

We have covered a lot so far, from raw pitch traces, through frequency and time normalization, to fitting, diagnosing, comparing, and interpreting GAMs, including some of the mess that comes with real data, like pitch-tracking artifacts and a borderline model comparison. That mess was not incidental to the workshop. Real data is rarely clean, and part of what GAMs offer is a principled way to work with that complexity rather than around it.

If you take away one thing, let it be this: a smooth term’s summary line is a starting point for investigation, not a final answer. Always look at the shape of what you fitted, not just the numbers describing it.

The material, dataset, and code remain freely available at the workshop repository for you to revisit and adapt at your own pace. If questions come up as you apply this to your own data, feel free to reach out. We hope this is a starting point for using GAMs in your research, not an endpoint.