3 ms·

> “We have a system currently that is almost entirely > reactive, a system influenced by anecdote and emotion,” > said Harris, who has dubbed her database the

> “We have a system currently that is almost entirely > reactive, a system influenced by anecdote and emotion,” > said Harris, who has dubbed her database the “Open > Justice” initiative. “The beautiful thing about numbers is > that they don’t lie.” Perhaps I'm being a bit contrarian, but numbers and statistics can lie more than you'd expect. Very few laymen can interpret statistics correctly, and you can even skew the proper interpretation by carefully selecting certain statistics which support your viewpoint. I would be careful in saying that numbers cannot lie, or that they don't lie. Meaningful metrics don't lie, but this is not the same thing (and even then, how do we define meaningful?). I wish people would consider that it is perfectly possible to lie using math, especially so when the average person does not maintain an advanced understanding of statistics.

Frondo 11y ago
"Very few laymen can interpret statistics correctly" -- Correct, and that's why there are people whose job it is to do so. I imagine the FBI, like any large organization, has at least a few of these people on the payroll. I don't imagine anyone who's enough of a data wonk to say something like ".. the beautiful thing about numbers .." is one of those untrained laymen. :)
kbenson 11y ago
Numbers don't lie. We lie as we try to make sense of the numbers and format them for display. At least when you have actual numbers underneath someone else can look and try to confirm or refute the prior presentation,or look for an entirely new meaning. Anecdote and popular sentiment to not allow this. It's not that having numbers means they can't be massaged to tell a specific story, but at least there's more than blind luck that goes into whether the conclusion makes sense.
- michaelbuddy 11y ago
  agreed, often you don't even need to massage too. If you collect some data without other data, you can almost logically derive an argument that wouldn't even be reasonable had other data been collected that puts it in perspective. So not only can you lie with data, complete data of a narrow query will lie to you even when you don't want it to.
- danso 11y ago
  I completely agree. Case in point, look at the front pages two most prominent examples of independent tracking of police shootings: The Guardian's "The Counted" reports "891 people killed this year [by police in the U.S.] - http://www.theguardian.com/us-news/ng-interactive/2015/jun/01/the-counted-police-killings-us-database http://www.theguardian.com/us-news/ng-interactive/2015/jun/0... The Washington Post reports: "758 people shot dead by police this year" - http://www.washingtonpost.com/graphics/national/police-shootings/ http://www.washingtonpost.com/graphics/national/police-shoot... The disparity is huge, but both sources could (and likely are, to the best of their ability) be reporting the truth. But they both have different opinions on what part of the issue should be prioritized. I believe the WaPo has chosen its limited methodology (which would exclude Freddie Gray, among others) for an understandable reason: police shootings are hard enough to classify, nevermind all police killings. Read the Guardian's methodology to see the pitfalls of their own approach: http://www.theguardian.com/us-news/ng-interactive/2015/jun/01/about-the-counted http://www.theguardian.com/us-news/ng-interactive/2015/jun/0... The Dallas PD is one of the few agencies that is actually publishing their officer-involved-incidents in a systematic, machine-readable way [1]...but if you check their official homepage and read a few of the narrative documents [2], you'll be able to see why things aren't cut and dry...and that how you choose to count things has a very strong impact on the numbers you end up with. [1] https://www.dallasopendata.com/api/views/4gmt-jyx2?accessType=DOWNLOAD https://www.dallasopendata.com/api/views/4gmt-jyx2?accessTyp... [2] http://www.dallaspolice.net/ois/ois.html http://www.dallaspolice.net/ois/ois.html

Frondo 11y ago

"Very few laymen can interpret statistics correctly" -- Correct, and that's why there are people whose job it is to do so. I imagine the FBI, like any large organization, has at least a few of these people on the payroll. I don't imagine anyone who's enough of a data wonk to say something like ".. the beautiful thing about numbers .." is one of those untrained laymen. :)

kbenson 11y ago

Numbers don't lie. We lie as we try to make sense of the numbers and format them for display. At least when you have actual numbers underneath someone else can look and try to confirm or refute the prior presentation,or look for an entirely new meaning. Anecdote and popular sentiment to not allow this. It's not that having numbers means they can't be massaged to tell a specific story, but at least there's more than blind luck that goes into whether the conclusion makes sense.

michaelbuddy 11y ago
agreed, often you don't even need to massage too. If you collect some data without other data, you can almost logically derive an argument that wouldn't even be reasonable had other data been collected that puts it in perspective. So not only can you lie with data, complete data of a narrow query will lie to you even when you don't want it to.
danso 11y ago
I completely agree. Case in point, look at the front pages two most prominent examples of independent tracking of police shootings: The Guardian's "The Counted" reports "891 people killed this year [by police in the U.S.] - http://www.theguardian.com/us-news/ng-interactive/2015/jun/01/the-counted-police-killings-us-database http://www.theguardian.com/us-news/ng-interactive/2015/jun/0... The Washington Post reports: "758 people shot dead by police this year" - http://www.washingtonpost.com/graphics/national/police-shootings/ http://www.washingtonpost.com/graphics/national/police-shoot... The disparity is huge, but both sources could (and likely are, to the best of their ability) be reporting the truth. But they both have different opinions on what part of the issue should be prioritized. I believe the WaPo has chosen its limited methodology (which would exclude Freddie Gray, among others) for an understandable reason: police shootings are hard enough to classify, nevermind all police killings. Read the Guardian's methodology to see the pitfalls of their own approach: http://www.theguardian.com/us-news/ng-interactive/2015/jun/01/about-the-counted http://www.theguardian.com/us-news/ng-interactive/2015/jun/0... The Dallas PD is one of the few agencies that is actually publishing their officer-involved-incidents in a systematic, machine-readable way [1]...but if you check their official homepage and read a few of the narrative documents [2], you'll be able to see why things aren't cut and dry...and that how you choose to count things has a very strong impact on the numbers you end up with. [1] https://www.dallasopendata.com/api/views/4gmt-jyx2?accessType=DOWNLOAD https://www.dallasopendata.com/api/views/4gmt-jyx2?accessTyp... [2] http://www.dallaspolice.net/ois/ois.html http://www.dallaspolice.net/ois/ois.html