Reported impressive reduction in accidents due to ADS-B In

[PeterNSteinmetz]

You want to ascribe meaning to a meaningless statistic, or insist that it is "interesting", I can't help that. That's why we have statistical analysis in the first place. So we don't accidentally ascribe meaning to data that aren't meaningful.

Why this strange insistence on disagreeing with what I have written?

The differences in the estimates of the rates of mid-airs are real differences in the observations. They have meaning if you were forced to bet on what the rates are, but that difference does not reach the level of statistical significance. The differences in the observed rates are not a "meaningless statistic". That is an over-simplification.

If one really wants to understand how to best interpret this test, rather than paying attention only to the significance level, pay attention to both sides of its performance. What is the power of the test to detect a difference? That can be determined by doing some research and making a computation and that will allow a better level of discussion.

Here is a reference about the power of a test for those who may be interested (versus just arguing endlessly). It explains the factors that affect the power of a test and how that is one aspect of a test's performance. https://stattrek.com/hypothesis-test/power-of-test.aspx

If someone wants to look up what the number of operations likely being used in the denominators of the ratios are (in the reference the report gave), I am happy to help compute the power of a binomial test to detect a difference in the rates.

 

[PeterNSteinmetz]

I think the fact that there were zero ADS-B midairs, and that there's still not a statistically significant difference from non-ADS-B midairs, shows that the test is not powerful enough.

My own intuition agrees that this may well be a case of low power due to a very low rate of mid-airs. But I don't feel that confident about it. There are likely a very large number of operations and so the denominator and number of observations would be quite large. It is possible that there is sufficient power in the test to detect any meaningful difference between the rates which would then argue that there is no real difference. Close call and I would have to actually see the numbers to have a better opinion. I'll run the power calculation if you will find the denominators

 

[PeterNSteinmetz]

I have no idea what the topic is anymore ...lol

I must say, I have been amused to observe the drift on threads I have started. Fun to see where things lead.