4 ms·

>Making a terse sociological claim from statistics is almost always open to interpretation. Almost, but in specific cases like above, it is not open to interpr

by TA0x0 6y ago

>Making a terse sociological claim from statistics is almost always open to interpretation. Almost, but in specific cases like above, it is not open to interpretation. It's a raw fact. >Statistics is the science of interpreting discrete data points, looking for patterns and relations. And organizing and presenting discrete data, without interpolation. >The only thing that stands alone without interpretation in statistics is the input data itself, and that tells us nothing. Wrong. Basic presentation without interpolation is a core pillar of statistics. >I suspect that for the specific claim you've made, the evidence supports the claim, but I'd want to see whether they've controlled for conflating the definition of "crimes committed" with "charged with crimes" or "convicted of crimes" to be certain. Even with controlling, African Americans are charged, commit, and are convicted of more crimes than Asian Americans. >Or whether the statistics are using the Bayesian or frequentist approach (after all, the claim being made is "are more likely," and that always assumes a giant pile of unstated priors). None of the above, just basic summary statistics. >Even still, whether the statistical claim being made is fact is uninteresting. If one assumes it's fact, the interesting question is "What do you do with it?" It's super interesting, considering what the OP asked. >Nothing. It's not nearly enough info to form theory or policy. It's more than enough info to form theory, and possibly policy. >Which is why my original response to your original post was about interpretation, not fact. And what I've stated above is fact, not interpretation. Back to square one. >Interpretation is where things get interesting. Facts without interpretation are dead as rocks. Wrong. Facts without interpretation are as alive as can be, and stand on their own.

shadowgovt 6y ago
> It's more than enough info to form theory, and possibly policy. Can you give an example? What kind of policy would you form around that statistic?
- TA0x0 6y ago
  >Can you give an example? Theory: there are manifestations in African American culture that breeds violence. Policy: we need to funnel anti-violence education funds towards African American school populations.
  - shadowgovt 6y ago
    Yeah, that's exactly why that's not enough data to put together policy. You might as well measure bumps on people's heads to predict whether they want anti-violence funding; you'll have about as much luck as you will chasing it after skin color or "manifestations in African American culture that breeds violence" (what does that even mean?).
    - TA0x0 6y ago
      >Yeah, that's exactly why that's not enough data to put together policy. Except it is, and I just proved it above. >You might as well measure bumps on people's heads to predict whether they want anti-violence funding Bumps on people's heads have nothing to do with violence. This is a non sequitur. >you'll have about as much luck as you will chasing it after skin color That makes no sense. The fact that the group is more violent is not open to interpretation, or "chasing", as I proved above. >manifestations in African American culture that breeds violence" (what does that even mean?). https://www.merriam-webster.com/dictionary/manifestation https://www.merriam-webster.com/dictionary/manifestation
      - shadowgovt 6y ago
        > Bumps on people's heads have nothing to do with violence. This is a non sequitur. I submit to you it's exactly as irrelevant as skin color. I submit to you if you take the population that you have described, change only the category from "African American" to another category and nothing else, and kept the same population of people with no other parameters changed, you would not see a change in the violence rate. Believing that measurable correlation implies causation is one of the major failings that I see often encouraged in the computer science space. It underpins some of the greatest failings of machine learning. You made claims about population and then claims about interventions that suggest that the population membership is causal. What if the causal issue is poverty? Poverty in the United States is so deeply correlated with being a member of a racial category that it is extremely hard to disambiguate effects that show up in race from effects that show up in income and asset levels. This is what I mean when I say lies, damn lies, and statistics. Because if the cause is poverty and not "African American culture," then violence interventions for African Americans are like "get your energy level up" interventions for people who are starving. You'll waste good money treating the wrong problem. You'll also miss people that also have the same problem but don't get any intervention because they don't fit the poorly-defined template.

by TA0x0 6y ago

shadowgovt 6y ago

> It's more than enough info to form theory, and possibly policy. Can you give an example? What kind of policy would you form around that statistic?

TA0x0 6y ago
>Can you give an example? Theory: there are manifestations in African American culture that breeds violence. Policy: we need to funnel anti-violence education funds towards African American school populations.
- shadowgovt 6y ago
  Yeah, that's exactly why that's not enough data to put together policy. You might as well measure bumps on people's heads to predict whether they want anti-violence funding; you'll have about as much luck as you will chasing it after skin color or "manifestations in African American culture that breeds violence" (what does that even mean?).
  - TA0x0 6y ago
    >Yeah, that's exactly why that's not enough data to put together policy. Except it is, and I just proved it above. >You might as well measure bumps on people's heads to predict whether they want anti-violence funding Bumps on people's heads have nothing to do with violence. This is a non sequitur. >you'll have about as much luck as you will chasing it after skin color That makes no sense. The fact that the group is more violent is not open to interpretation, or "chasing", as I proved above. >manifestations in African American culture that breeds violence" (what does that even mean?). https://www.merriam-webster.com/dictionary/manifestation https://www.merriam-webster.com/dictionary/manifestation
    - shadowgovt 6y ago
      > Bumps on people's heads have nothing to do with violence. This is a non sequitur. I submit to you it's exactly as irrelevant as skin color. I submit to you if you take the population that you have described, change only the category from "African American" to another category and nothing else, and kept the same population of people with no other parameters changed, you would not see a change in the violence rate. Believing that measurable correlation implies causation is one of the major failings that I see often encouraged in the computer science space. It underpins some of the greatest failings of machine learning. You made claims about population and then claims about interventions that suggest that the population membership is causal. What if the causal issue is poverty? Poverty in the United States is so deeply correlated with being a member of a racial category that it is extremely hard to disambiguate effects that show up in race from effects that show up in income and asset levels. This is what I mean when I say lies, damn lies, and statistics. Because if the cause is poverty and not "African American culture," then violence interventions for African Americans are like "get your energy level up" interventions for people who are starving. You'll waste good money treating the wrong problem. You'll also miss people that also have the same problem but don't get any intervention because they don't fit the poorly-defined template.