Life distribution with censored data that is difficult to fit a distribution model

[#1]

Hi SAS community

We have a set of data 12 censored interval data points: (0-30, 0-30, 0-30s, 0-30, 0-30, 0-30, 0-30, 0-30, 0-30, 0-30, 89-119, 97-127)minutes. The data points indicates the time interval in which a failure of our product was observed. We want to calculate the time at which 90% of our products are failed (target is below 600minutes). Usually we use the life distribution menu to fit a model to our censored data. We choose the best fitted distribution and use custom estimation to calculate the confidence interval for the failure probability of 0.90. Normally it works perfectly fine because the censored data intervals are more spread, but this dataset does not allow for a good fitted distribution. Hence we get a very wide 95% confidence interval for 0.90 failure probability. Any suggestion on how to get around this? 

[#2 · Accepted solution]

Limited data cause poor performance (wide interval estimates). There is no way around the wide intervals except to collect more data.

It is best to use as much scientific and engineering knowledge about the failure mode as possible when selecting the model instead of relying on a sample statistic or a model criterion. This reliance is even riskier when the data are so limited. The probability plot exhibits poor support from the data for this model.

[#3]

Hi Mark

Thank you for your help. 

Kind regards
Thor