Life distribution: The failure rate/probability plot does not match with the table

[#1]

Hello,

I am new to the JMP life distribution plot.

I think it is a basic understanding issue.

The below is the test table I made.

According to the description: https://www.jmp.com/support/help/en/18.1/index.shtml#page/jmp/event-plot.shtml#ww1386833

I think it meas there are total 100 samples.

And exactly at 4 readings: 50, 100, 150 and 200, there are 25 new failures happened.

So the failure rate I expected will be:

50, 25%

100, 50%,

150, 75%

200,100%.

Then I lauch the life distribution with setting as below:

However, the dots in the Weibull plot are very different from the failure rate I expected.

For example, the first read point at time 50 has failure rate as ~13%.

I don't understand why.

Could you help me with this?

Thank you very much for your help!

I am using JMP17

[#2]

I think what really confuses me is that JMP provided the left-right censor capability to indicate when the failures happened.

So if this is not an insitu monitored test, I agree with you that we cannot say exactly at t=50h, there are 25% failures happened at the same time, but instead we can only say 25% of the population fails between t=0 and t=50. Then the way JMP handles the probability calculation and the weibull fit makes sense.

But now I have insitu monitored test and I know exactly when each of the failure happened. 

But JMP still interpreted the data like I don't have an insitu monitoring capability by doing some estimates to adjust the failure rate.

I don't really understand the reasoning behind this.

[#3]

This pdf document from Discovery 2017 provides a good overview of how to analyze various kinds of censoring in JMP, and the Maximum Likelihood Estimates (MLE) associated with them.  The one thing it omits is Exact Data, but you've already seen how to enter that with your original example (Start time = End time).  As demonstrated in the document, JMP allows you to enter mixed censoring types in your data table.  The MLEs are calculated differently for each type, so yes - it certainly matters and you should always strive to have honest, representative data (incl. censoring) to get the most meaningful fits. 

[#4]

some details about probability calculation in Life Distribution | different versions:
Life distribution interval censoring compare distributions plot