nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Probability from Residual Probability Plot in the Parametric Survival Fit Platform

MHz

Hi everyone,

I would like to know more about the Y-axis in the Residual Probability Plot from the Parametric Survival Fit Platform.

It is possible to save the Fitted Failure Probability in a new column. Is it possible to save the unfitted Failure Probability? Another way to say it: I would like to save the failure probability of the experimental point.

If it's not possible, does anyone knows the formulas, when censor is part of the data?

Thanks,

Accepted answers

peng_liu

The Y-axis in the residual probability plot is a probability scale axis. E.g. if you fit a Weibull distribution, the Y-axis is a Weibull scale axis; if you fit a Lognormal distribution, the Y-axis is a Normal scale axis. The residual probability plot is set up by such, so that if the assumption the model holds, the residuals should appear roughly in a straight line.

The x-position of the markers are computed in a way as in my answer to your other post Re: Standardized residuals from Parametric Survival Fit Platform

The y-positions of the markers does not have an explicit formula in general if censoring type is not simple. But think y-positions as non-parametric estimates of the distribution at those points.

For your second question, I guess that you want to calculate failure probability of an arbitrary setting at an arbitrary time, which is not in the data table. If my understanding is right,you can choose one of the following:

1) save the probability formula, add a new row with your covariate setting and a time value. Read out the number at that row in the formula column.

2) use the menu "Estimate Survival Probability" in the platform to calculate directly, by giving your settings and time. See results in the report after committing the setup.

Notice both calculate probabilities at given times, so censoring is irrelevant.

All comments

Mark_Bailey

The probability of the data (i.e., life time) is obtained from the likelihood function, l( y | model ). It is the conditional probability that is estimated with model using the maximum likelihood estimates of the model parameters. The data does not have a probability as such. On the other hand, you could use the non-parametric estimate (i.e., Kaplan-Meyer product moment) for the marginal probability. You could also assume a distribution model for parametric estimates.

See Reliability and Survival > Life Distribution. The K-M estimates are presented automatically.