Thursday, February 2, 2017

America in crisis: Reality or Manipulated Perceptions?

Captioned "Racist, homophobic graffiti was spray painted at Withrow University High School this past weekend. "  Source
H/T to BigHeadRob on the source.  The article is much better than the photo caption.
The reporter sees sinister signs of a fragmenting society.  I see sloppy, sophomoric attention whoring.

If this was truly an effort on the part of the vast, Alt-right to intimidate the left, don't you think they would have sent somebody who actually knew how to draw a swastika?   It was on the sidewalk at a high school.  It is a safe bet that it was done by one of the students who attended that school or one of their cross-town rivals.

So how can one discern whether we have in fact become less civil or if it is manufactured "media hype" and social media "reverb"?
In a word:  Statistics and history.
Data from the FBI Uniform Crime Statistics web page.  Aggravated Assault chosen as the most representative and best populated metric for "un-civil" behavior.  The mean for the period from 2001-2015 is 273/100,000 and 1.28 standard deviations is 37.2
Data from the FBI Uniform Crime Statistics web page.  Rape chosen as the most representative, and best populated metric for "violence toward women".  The mean for the period from 2001-2015 is 29.8/100,000 and 1.28 standard deviations is 3.23
The significance of 1.28 standard deviations is that the (mean + 1.28 standard deviations) defines the 90% cumulative for a normal (or bell) curve.  That is, if you were given an additional piece of data and it was above the mean + 1.28 standard deviations you could say with 90% certainty that something had changed in the process and that there is only a 10% chance that "random variation" generated that "flier".

The time span was chosen to represent a near equal number of years when the President was Republican (Bush) and Democrat (Obama) to wash out any reporting (or other) bias.

The effect of sample size
One reason that the FBI is loath to report monthly numbers is that random noise, while dampened in large samples, is amplified in small samples.

Consider a small city.  It might not have any homicides for twenty years...and then have three in one month.  Just random noise.

Decreasing the size of the sample to 1/4 of the original sample size results in a doubling of the apparent "noise".  Decreasing the size of a sample to 1/12 of the original can be expected to result in SQRT(12) times increase in variation, or to increase the standard deviation by about 3.5

Grinding the numbers, an observer can make a statistically defensible conclusion about loss of civility if the Annualized Aggravated Assault rate for January 2017 is 402/100,000 (or higher) or if the Annualized "legacy definition" Rape rate for January 2017 is 41/100,000 (or higher).

Rates that are less than 402 and 41 respectively are not high enough to make any statement with 90% certainty.

"News" that trumpets some minor up-blip fail the a priori requirements of good statistics.

I wait with bated breath to see the January 2017 data.

---Note added later:  I don't want to give the impression that I believe that the world is entirely safe and benign.  I believe that between 5% and 10% of the population suffer from EDD, that is Empathy Deficiency Disorder.  They have few inhibitions against doing violence to others.  My argument is that that percentage of people with EDD did not jump up when Donald Trump was elected.---

4 comments:

  1. That is the last thing the emoticons want is someone using actual facts... Doesn't fit the agenda, even with the error bars. And concur, monthly reports are pretty much useless!

    ReplyDelete
    Replies
    1. It is not that monthly reports are useless, it is that most people try to read more into them than is actually in the data. They want to make a stink about a 5% or 10% up-tick when they need about a 30% up-tick to be significant.

      I wish I had the FBI data per-100,000 data broken down by month for Jan-2001 through most recent. It would also be grand if it were broken down by urban/non-urban counties. I would have a lot more confidence about making 90% confidence guesses.

      Delete
    2. Understood, but you and I both know that data isn't available on a timely basis.

      Delete
  2. Also, I see them post numbers for cities like Chicago, from last year in January, and then this year in January, and then extrapolate that somehow homicides are up so many %. Like you say, totally meaningless numbers. Taken as a running total of a larger number of months, or as a year to year comparison, it might have some significance. Done like this, as is the way that I see some in the media, both the right and the left, doing it, and the only thing that it proves is that the one using the statistics is trying to prove something and is cherry picking numbers in order to do so. Wait, I have heard the climate scientists accused of doing the same thing. Must be a coincidence.

    ReplyDelete