Use Nonparametric Distribution Analysis (Arbitrary Censoring) to estimate the reliability of a product when you have arbitrarily-censored data and no distribution fits your data. Arbitrarily-censored data include left-censored observations and/or interval-censored observations. For more information, go to Data censoring.

Depending on which nonparametric method you choose, you can do the following:

- Estimate the percentage of items that will fail or survive at various time intervals
- Display hazard plots and survival plots to view the failure and survival probabilities
- Test the equality of the survival curves for multiple samples

You can also use a nonparametric distribution analysis to validate and compare the results from a parametric distribution analysis. A nonparametric distribution analysis is less efficient than a parametric analysis with an appropriate distribution model, and thus often produces wider confidence intervals. However, because a nonparametric analysis does not require fitting any distribution to your data, its results will not be adversely affected by a poor distribution fit.

When your product fails in different ways, use a failure mode analysis to evaluate the impact of each type of failure on the overall reliability. Each failure mode is assumed to be independent. By analyzing each failure mode individually, you can more easily prioritize your improvement efforts.

To perform a nonparametric distribution analysis for arbitrarily censored data, choose

.- If your data contain only exact failure times and/or right-censored observations, use Nonparametric Distribution Analysis (Right Censoring).
- Parametric methods provide more precise results and estimate more types of functions. Therefore, if you can fit a parametric distribution to your data, use Parametric Distribution Analysis (Arbitrary Censoring).