Survival analysis manages the reality of 'censored' data—records where the event of interest hasn't occurred by the end of your study. Ignoring these incomplete records leads to skewed averages. Lattice approaches this through a three-stage methodology: first, the LLM parses your dataset to identify time and event columns, ensuring explicit confirmation of binary event encoding. Second, the deterministic engine executes specialized statistical libraries to handle non-parametric curves or semi-parametric regression. Finally, the LLM interprets the coefficients, hazard ratios, and p-values into plain language, helping you decide whether to publish, adjust for covariates, or re-examine your model based on proportionality assumptions.
When to choose this family
- You have time-stamped data and need to calculate how long until an event occurs.
- You need to compare two or more groups (e.g., treatment vs. control) despite having incomplete or censored observations.
- You want to determine which variables independently influence the hazard rate of an event occurring.
- You require publication-ready survival curves that include at-risk tables and confidence intervals.
What this family does
This family quantifies the duration until a specific outcome, such as machine failure, patient recovery, or account cancellation. It accounts for censoring, meaning it mathematically incorporates subjects who dropped out or reached the study's end without the event happening, rather than discarding them.
Tools here provide both visual and numerical insights, from generating Kaplan-Meier plots to identifying independent risk factors using Cox regression. The results include key metrics like median survival time and hazard ratios, giving you a clear view of how different factors drive event timelines.
Differentiation from other methods
Unlike standard regression models that predict a simple output value, this family focuses on the probability of an event occurring over time. It is distinct because it handles 'censored' data, which traditional time-series or standard linear models treat as missing data, leading to significant bias.
While standard regression assumes the same impact from variables across the board, survival models test proportionality. This ensures that the influence of a variable—like a patient's age or a component's batch—remains consistent throughout the observation period.
Common pitfalls
A common mistake is failing to verify the event encoding. Lattice requires clear identification of which value represents the event and which represents censoring; assuming defaults can flip your results entirely.
Another risk is ignoring the proportionality assumption in regression. If the effect of a variable changes over time, a standard Cox model may report misleading results. The analysis tools in this family provide automated diagnostics to alert you when these assumptions are violated so you can adjust your strategy.
Frequently asked questions
- How do I know if my data is censored?
- Censored data occurs when you know a subject hasn't experienced the event yet (e.g., they are still using your product or are still alive at the study's end). When you use Lattice's survival tools, the LLM will explicitly ask you to define which column represents time and which binary column marks the event (1) vs. censoring (0).
- What should I do if my model reports a 'proportional hazard violation'?
- This means your variables don't have a constant effect over time, violating a key assumption of the Cox model. Lattice tools will flag this using Schoenfeld residual tests. You should look for our suggestions to use stratification for those specific variables or consider alternative modeling approaches mentioned in the tool's diagnostic output.