This person is older than usual?

monnettohio

Today I got this warning: "This person's birth and death dates indicate death at age 93, which is older than usual." So we are going to redflag people's profiles because of acceptable longevity when they are less than 100? And how does one correct this when it is accurate? Seems a bit like dooming the profile to LOW unless someone fiddles with a date

This seems a bit much, especially when the birth and death are documented with sources. This person was born in 1909 and died in 2003. Nonagenarians are not uncommon in today's world as they were in the early 1960s when my great-grandmother made the newspaper for making it to 97.

I could understand getting this IF the person were listed at an extreme, nearly impossible age at the date of death, say 125 years old - that is a red slag. But Jean Calmont lived to be 122 and is the oldest person ever documented from recorded birth, life, and death. So its possible, but it is rare.

Gordon Collett

Is this particular flag dismissible? I don't have an example so I can't check. Try clicking on the "This person's birth…" statement and see if a dismiss box comes up.

monnettohio

Here is an example: Harriett Belle Jones, age 97, https://www.familysearch.org/tree/person/details/KZXH-293

Here is another example, Oliver Hamilton, age 104: https://www.familysearch.org/tree/person/details/9H6L-NMF

Frankly, I can understand that this flag kicks in at 100 or 105. There are places, people, and family groups where longevity is a simple fact. And we live in an era when we have more nonagenarians and centenarians than ever before.

But this flag, for me, is not helpful IF the sources have death records attached to the subject. After that, it is simply an annoyance.

Gordon Collett

Checking your examples, that flag is dismissible. Simply click on the text of the flag then click Dismiss.

There will always be flags that question correct data. And there will always be need to balance the strictness of the flag based on time and place and to balance making the flag useless because it is too lax against making it useless because it is too strict. That is why the ability to dismiss flags against data that has been confirmed to be correct is valuable.

If you never have to dismiss a flag, it is probably too lax. If you always have to dismiss a flag is it clearly too strict. If you have to dismiss a flag on one out of a hundred people, it's probably working pretty well.

So just dismiss the flags on Harriett and Oliver then keep track of how often you need to do this on others. If you find that you are dismissing this flag on more than say 10% of your family, then come back and complain again.

Rhonda Budvarson

@monnettohio thank you for your feedback. The average age of death for 2003 was 77 years of age. So your person was indeed older than the norm. Now that you have checked the facts to insure that this is correct, please dismiss the issue by clicking on the notice and clicking the dismiss button.

Chas Howell

As Gordon Collett rightly points out, there is a balancing act needed. In this case, I agree with monnettohio, in my opinion, a flag for a 93 year old is too strict. The ratio of how often you have to dismiss a flag to the total number of people is a good gage, but I think another criterion for judging the appropriateness of a flag is to ask, how often does the flag actually catch an error. About 5% of the people in the US will reach 90 years old. How many of those flags for ages 90 to 99 actually catch a data error for that age group? A flag will lose its effectiveness if it is in error too often.

Gordon Collett

@Chas Howell Good point. If the routine cries "wolf" too often, users may just decide to just ignore it altogether.

However, here the concepts of precision and specificity come into play. With any type of test, the test can have high or low precision and high or low specificity. These two play off each other and result in all tests having the following:

1. True positives - tree profile has death at age 95 but birth or death date is wrong - person died at 77.
2. False positives - tree profile has death at age 95 and birth or death date is correct - person died at 95
3. True negatives - tree profile has death at age 77 and birth or death date is correct - person died at 77
4. False negatives - tree profile has death at age 77 but birth or death date is wrong - person died at 95

The usual goal is to have the maximum number true positives and the lowest number false negatives.

If the death date cut off was set to 120, the flag "died too old" would always be right. It would always be a true positive and you would never have a false positive. But you would have a unacceptable rate of false negatives because no one with a questionable death date would ever be flagged which would make the flag useless.

Determining the age cut off that gives a good balance between true positives and false negatives is the hard part.

I guess one way to judge the appropriateness of the current flag setting is to look at a huge set of profiles in Family Tree of people born in the US during a set time period that have both a birth and death date and see how many have a death date showing they reached 90. If that is about 5%, then the flag could probably be set higher. If it is 25%, then the flag is probably good where it is.

@Rhonda Budvarson Could you clarify your comment? I do hope the average death age of 77 is not what is being used to set the flag. That would mean that 50% of the time the flag would be set for died too old because average means that half of all people died after age 77. I would have assumed that they are using something like the fairly common two standard deviations above the mean as the cut off point.

Rhonda Budvarson

@Chas Howell @Gordon Collett Thank you for your input and comments. A ticket has been issued for this issue. I will ask for the specific year range being used.

roberthparker3

@Chas Howell we are currently using the 90th percentile age at death, which is 88 years old for people born in the United States 1800-1920 birth segment. The 99th percentile for the same segment is 98 years old.

I like the idea of not flagging if there is a source involved. I also like the idea of increasing the threshold to reduce the noise.

Gordon Collett

Two standard deviations from the mean is typically about the 95%tile. There must be some good reasons why statisticians like that spot so maybe using that value, which would probably be about 93 to 95 years old, could be looked at.

(My last statistics class was far to long ago to remember why 2σ is used. I just remember having to calculate it too many times.)