RSS
From the New York Times Upshot section:
Murders Are Rising. Blaming a Party Doesn’t Add Up.
The data reveals a different picture than the party-driven explanation President Trump and the Department of Justice have offered.By Jeff Asher
Sept. 28, 2020Jeff Asher is a crime analyst based in New Orleans and co-founder of AH Datalytics. You can follow him on Twitter at @Crimealytics.
By the way, Mr. and Mrs. Asher have infant triplets.
… Over all, in 59 cities with murder data available through at least July this year, murder is up 28 percent relative to the matching time frame in 2019. …
Big cities tend to overstate national crime trends, so a smaller rise in murder would be expected nationally, but a 15 percent increase in murders nationally in 2020 would be the largest one-year increase in modern American history in terms of both raw numbers and percent change (reliable data on national murder trends began in 1960).
Murder started rising after the emergence of Black Lives Matter at Ferguson in August 2014, but then dropped a little after Trump’s election, only to surge after BLM returned with a vengeance in the Summer of George.
When Black Lives Matter is riding high, blacks shoot each other more.
Here’s my sort of Jeff Asher’s spreadsheet of murders in 59 reporting cities through various dates in September compared to 2019. Some interesting cities, such as Pittsburgh and Memphis, are missing, but overall it’s a pretty good tabulation.
| City | 2019 | 2020 | Change |
| All Reported Cities | 3905 | 5003 | 28% |
| Lubbock | 9 | 22 | 144% |
| Milwaukee | 66 | 134 | 103% |
| Fort Wayne | 13 | 24 | 85% |
| Minneapolis | 32 | 57 | 78% |
| Riverside | 7 | 12 | 71% |
| Louisville | 62 | 104 | 68% |
| Toledo | 22 | 36 | 64% |
| Fort Worth | 31 | 49 | 58% |
| Colorado Springs | 14 | 22 | 57% |
| Lexington | 16 | 25 | 56% |
| Aurora | 15 | 23 | 53% |
| New Orleans | 87 | 133 | 53% |
| Boston | 30 | 45 | 50% |
| Chicago | 374 | 560 | 50% |
| Denver | 43 | 63 | 47% |
| Arlington | 13 | 19 | 46% |
| Austin | 23 | 33 | 43% |
| Atlanta | 69 | 97 | 41% |
| San Bernardino | 25 | 35 | 40% |
| New York | 236 | 327 | 39% |
| St Louis | 136 | 185 | 36% |
| Nashville | 54 | 73 | 35% |
| Cincinnati | 54 | 73 | 35% |
| Greensboro | 30 | 40 | 33% |
| Indianapolis | 98 | 129 | 32% |
| Houston | 176 | 229 | 30% |
| Philadelphia | 243 | 315 | 30% |
| Cleveland | 88 | 114 | 30% |
| Miami | 60 | 77 | 28% |
| Sacramento | 18 | 23 | 28% |
| Kansas City | 113 | 144 | 27% |
| Tulsa | 37 | 47 | 27% |
| Oakland | 52 | 66 | 27% |
| Seattle | 23 | 29 | 26% |
| San Antonio | 66 | 83 | 26% |
| San Diego | 24 | 30 | 25% |
| Detroit | 187 | 230 | 23% |
| San Francisco | 23 | 27 | 17% |
| Jacksonville | 93 | 108 | 16% |
| Charlotte | 73 | 84 | 15% |
| Portland | 22 | 25 | 14% |
| Los Angeles | 197 | 223 | 13% |
| Washington | 126 | 140 | 11% |
| Stockton | 28 | 31 | 11% |
| Bakersfield | 27 | 29 | 7% |
| Long Beach | 18 | 19 | 6% |
| Omaha | 23 | 24 | 4% |
| Dallas | 151 | 157 | 4% |
| Las Vegas | 64 | 64 | 0% |
| Lincoln | 4 | 4 | 0% |
| Plano | 1 | 1 | 0% |
| Baltimore | 247 | 240 | -3% |
| San Jose | 24 | 21 | -13% |
| Newark | 37 | 31 | -16% |
| Durham | 22 | 18 | -18% |
| Oklahoma City | 39 | 31 | -21% |
| Virginia Beach | 9 | 6 | -33% |
| Chula Vista | 11 | 7 | -36% |
| Anchorage | 20 | 6 | -70% |
These are pretty awful numbers in BLM cities like Minneapolis (+78%), Milwaukee (+103%), Louisville (+68%), and Atlanta (+41%), slightly less awful in Antifa cities like Seattle (+26%) and Portland (+14%).
OT
Coronachan has just made 7 digits!
One million deaths!
On an Earth where 57 million die each year and 141 million are born each year.
Covid-19 pandemic has cost the world’s economy $3.8TRILLION ‘and made 147 million people unemployed’, study claims
Maybe it’s because the police aren’t shooting enough people?
Plano, IL? That is an interesting town to include. I think it might have 3,000 people. (ok I checked: 10,000) Home of Plano Molding, they make plastic ammo cans, deck cabinets, etc.
Baltimore actually went down. Protests here have been relatively muted though.
I wonder why murders are down slightly, 2020 cf. 2019.Replies: @stillCARealist, @Mark Roulo, @Tony, @Ga On My Mind
Don’t miss Ann Coulter in Takimag. Indispensable.
If you have red blood in your veins, this’ll make it boil.
https://www.takimag.com/article/innocent-until-proven-trump-supporter/
Meanwhile:
Maryland county to pay $20mn in ‘heinous,’ ‘brutal’ & ‘senseless’ shooting of HANDCUFFED black man by black police officerFollowed by heartfelt encomiums. There must be more to this story ... guess a black guy pissed off a black officer - open to more spontaneous responses - enough.Replies: @anon, @Cloudbuster, @TomSchmidt, @DCThrowback
Less boiling more leisure. The world is making it difficult to relax. Weed and headphones; just like PB&J.
Almost certainly Plano, TX, with a population of almost 260,000. Plano is a very affluent suburb of Dallas.
Seems like they shoot plenty of non-blacks too.
But for some reason this doesn’t trouble anyone.
Plano,Texas?…….
The (ideologically heightened) discord between police and urban communities is raising murder rates in Republican controlled cities, too, so can we just agree to not politicize the matter?
A tale of two crumbling cities. So, who’s the world’s worst mayor: New York’s de Blasio or London’s Khan? It’s a close-run race...The above needs to be widened to book size.
I other news, pubs in London must close at 22:00, except the one in the Palace of Westminster which has been classified as "workplace canteen". I wonder whether one has to wear a powdered wig and a Mouche?
Also, RT has now said at least twice that Trump has said the he wasn't going to "leave peacefully", something even Snopes classifies as "mostly false". Good job.
Anchorage ? and Alaska is such a rich source of whites behaving badly ……….still love Sarah.
I’m currently swamped, but if anyone could create a spreadsheet with the change in murder rate Steve computed vs % blacks and %vote for Hilary in 2016 it could be interesting.
If you have red blood in your veins, this'll make it boil.
https://www.takimag.com/article/innocent-until-proven-trump-supporter/Replies: @Charon, @El Dato, @SunBakedSuburb, @Glt, @Ben tillman
It seems as though every single (in)famous story used for MSM purposes has turned out to be a fraud, an outright lie, a hoax.
None of those terms is strong enough to capture the meaning of what’s going on now. Even “blood libel” isn’t enough.
And in case *that* weren’t enough, TPTB are doing everything they can to erase the truth wherever they find it. If they retake the Senate, sites (like this one and taki’s) which dare to expose the facts will be erased too.
This must be what it’s like to lose a war.
Why can’t we blame a political party? Seems like the mayor of each city should belong to one. I’m assuming that data was available but not to the authors liking.
2010 Census shows Anchorage is only 62% Non-Hispanic White. No doubt it’s become less White since then.
NYT and the rest of the MSM (which are fanatically anti-Trump) created moral panic and mass hysteria based on lies and that campaign of lies results in soaring homicide rates. Is pointing that out “politicization”?
The mayor’s party is in Asher’s spreadsheet. I left it out because I’m not that interested in it.
White queer/atheist/Jewish Antifa causes less murder than peaceful uber Numinous Negro Black Lives Matter.
Yes, racists, Negroes are superior at some things.
By my reckoning, Chicago wins, with Lubbock down in 11th place. I got that by sorting not by percent increase, but by number of standard deviations of increase based on a Poisson model (which seems appropriate). Thus one standard deviation is the square root of the mean (which I took to be the mean of murders in 2019 and 2020).
The last column, T, is the number of standard deviations:
T = (m2020-m2019)/sqrt((m2020+m2019)/2)
City m2019 m2020 T
Chicago 374 560 8.61
Milwaukee 66 134 6.80
New_York 236 327 5.42
Louisville 62 104 4.61
New_Orleans 87 133 4.39
Philadelphia 243 315 4.31
St_Louis 136 185 3.87
Minneapolis 32 57 3.75
Houston 176 229 3.72
Lubbock 9 22 3.30
Atlanta 69 97 3.07
Detroit 187 230 2.98
Indianapolis 98 129 2.91
Fort_Worth 31 49 2.85
Denver 43 63 2.75
Kansas_City 113 144 2.73
Toledo 22 36 2.60
Cleveland 88 114 2.59
Fort_Wayne 13 24 2.56
Boston 30 45 2.45
Nashville 54 73 2.38
Cincinnati 54 73 2.38
Miami 60 77 2.05
Lexington 16 25 1.99
San_Antonio 66 83 1.97
Colorado_Springs 14 22 1.89
Austin 23 33 1.89
Aurora 15 23 1.84
San_Bernardino 25 35 1.83
Oakland 52 66 1.82
Los_Angeles 197 223 1.79
Greensboro 30 40 1.69
Riverside 7 12 1.62
Tulsa 37 47 1.54
Jacksonville 93 108 1.50
Arlington 13 19 1.50
Charlotte 73 84 1.24
Washington 126 140 1.21
Seattle 23 29 1.18
San_Diego 24 30 1.15
Sacramento 18 23 1.10
San_Francisco 23 27 0.80
Portland 22 25 0.62
Stockton 28 31 0.55
Dallas 151 157 0.48
Bakersfield 27 29 0.38
Long_Beach 18 19 0.23
Omaha 23 24 0.21
Plano 1 1 0.00
Lincoln 4 4 0.00
Las_Vegas 64 64 0.00
Baltimore 247 240 -0.45
San_Jose 24 21 -0.63
Durham 22 18 -0.89
Newark 37 31 -1.03
Virginia_Beach 9 6 -1.10
Chula_Vista 11 7 -1.33
Oklahoma_City 39 31 -1.35
Anchorage 20 6 -3.88
Your comment made me curious, so I looked around and found this 2017 paper.
Monitoring Volatile Homicide Trends Across U.S. Cities
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2977556Note their method and conclusion.Any thoughts on using the Poisson vs. binomial distribution for modeling homicides? This is getting beyond my statistical pay grade, but it looks like the difference is small (in the limit they are the same) and if anything using Poisson understates the rarity of evens (mean = variance vs. mean > variance for binomial).
https://keydifferences.com/difference-between-binomial-and-poisson-distribution.html
They include a link to data and code for replication, but use SPSS which complicates things a bit for those who don't have it. This R package might be helpful.
https://www.r-bloggers.com/2013/11/spss-to-r-an-r-package-to-convert-spss-syntax/
I'm trying to work out how to interpret things given that they draw very different conclusions about the presence of a Ferguson effect based on what look like fairly similar numbers. Taking the Baltimore panel of Figure 4 as an example we see the 2014 homicide rate was about 33 compared to about 55 for 2015.
If I understand your method correctly this would indicate a (55 - 33) / sqrt((55+33)/2) = 3.3 SD change. That is about a 1 in 2000 event. Yet looking at their graphic we see 2015 barely falls outside their 50% prediction interval and is well inside their 80% prediction interval.
Those approaches give vastly different indicators of how rare an event the Baltimore 2014-2015 change in homicides was. Perhaps a useful check would be to apply your methodology to each yearly change and see how rare all of those look?
What do you (or anyone else who is statistically literate enough to understand all of this) think?Replies: @International Jew, @James B. Shearer
Also, using 2020 data to calculate the mean has the effect of understating the variance. Since what we're really interested in is assessing is just how big of an anomaly 2020 has been it I think it would have been better to just use 2019 as the mean.
I think this was the joker i heard on npr a few weeks back. The upshot is that the percent change in republican mayor cities vs. Democrat mayor cities was like 25 vs. 29 or something. Agree with Steve that it’s not interesting.
Two other tidbits that emerged from the interview that enraged me at the time were
(1) his “research” showed that less than 1 percent of police responses are to murders, which somehow proved that adding more police wouldn’t reduce murders. Hey nerd, what percentage of criminal complaints are murder? Less than 1 percent? My prior is not to trust a datalyst that can’t do basic Bayesian training…
(2) a big deal was made by the presentrix that he was former cia. Since when has that been the ne plus ultra of trustworthiness for the left wing?
The last column, T, is the number of standard deviations:
T = (m2020-m2019)/sqrt((m2020+m2019)/2)
City m2019 m2020 T
Chicago 374 560 8.61
Milwaukee 66 134 6.80
New_York 236 327 5.42
Louisville 62 104 4.61
New_Orleans 87 133 4.39
Philadelphia 243 315 4.31
St_Louis 136 185 3.87
Minneapolis 32 57 3.75
Houston 176 229 3.72
Lubbock 9 22 3.30
Atlanta 69 97 3.07
Detroit 187 230 2.98
Indianapolis 98 129 2.91
Fort_Worth 31 49 2.85
Denver 43 63 2.75
Kansas_City 113 144 2.73
Toledo 22 36 2.60
Cleveland 88 114 2.59
Fort_Wayne 13 24 2.56
Boston 30 45 2.45
Nashville 54 73 2.38
Cincinnati 54 73 2.38
Miami 60 77 2.05
Lexington 16 25 1.99
San_Antonio 66 83 1.97
Colorado_Springs 14 22 1.89
Austin 23 33 1.89
Aurora 15 23 1.84
San_Bernardino 25 35 1.83
Oakland 52 66 1.82
Los_Angeles 197 223 1.79
Greensboro 30 40 1.69
Riverside 7 12 1.62
Tulsa 37 47 1.54
Jacksonville 93 108 1.50
Arlington 13 19 1.50
Charlotte 73 84 1.24
Washington 126 140 1.21
Seattle 23 29 1.18
San_Diego 24 30 1.15
Sacramento 18 23 1.10
San_Francisco 23 27 0.80
Portland 22 25 0.62
Stockton 28 31 0.55
Dallas 151 157 0.48
Bakersfield 27 29 0.38
Long_Beach 18 19 0.23
Omaha 23 24 0.21
Plano 1 1 0.00
Lincoln 4 4 0.00
Las_Vegas 64 64 0.00
Baltimore 247 240 -0.45
San_Jose 24 21 -0.63
Durham 22 18 -0.89
Newark 37 31 -1.03
Virginia_Beach 9 6 -1.10
Chula_Vista 11 7 -1.33
Oklahoma_City 39 31 -1.35
Anchorage 20 6 -3.88Replies: @International Jew, @res, @silviosilver
Sorry about the lack of formatting. I had html tags but unz.com stripped them all out.
If you have red blood in your veins, this'll make it boil.
https://www.takimag.com/article/innocent-until-proven-trump-supporter/Replies: @Charon, @El Dato, @SunBakedSuburb, @Glt, @Ben tillman
Joggers control the USA. The deal is to jog along or get yourself drilled another asshole. But yes, GoFundMe should get the whole Gawker treatment and then some. If you have to “bake the cake” then you have to “present the page”.
Clearly indications that he was about to turn his life around.
Meanwhile:
Maryland county to pay $20mn in ‘heinous,’ ‘brutal’ & ‘senseless’ shooting of HANDCUFFED black man by black police officer
Followed by heartfelt encomiums. There must be more to this story … guess a black guy pissed off a black officer – open to more spontaneous responses – enough.
No body cam, no clear witnesses, and the officer isn't talking to the press (which is smart). But that won't prevent the press from trying him in absentia.
The criminal trial hasn't occurred yet. It will be interesting to see what comes out then.
Check out who runs PG county for a better explanation.
No.
A tale of two crumbling cities. So, who’s the world’s worst mayor: New York’s de Blasio or London’s Khan? It’s a close-run race…
The above needs to be widened to book size.
I other news, pubs in London must close at 22:00, except the one in the Palace of Westminster which has been classified as “workplace canteen”. I wonder whether one has to wear a powdered wig and a Mouche?
Also, RT has now said at least twice that Trump has said the he wasn’t going to “leave peacefully”, something even Snopes classifies as “mostly false”. Good job.
Isn’t “Upshot” an unfortunate header for this NYT dissertation?
https://www.nytimes.com/newsletters/upshot
Baltimore seems to be the outlier on this list. ~60% black, home of Freddie Gray and the mostly peaceful protests in his memory, very high baseline murder rate, early center of BLM activity.
I wonder why murders are down slightly, 2020 cf. 2019.
You can see the data here: http://mistybeach.com/mark/BaltimoreHomicides.html
"Baltimore ended 2019 with 348 homicides on record, according to data compiled by The Baltimore Sun. The year had already set a grim record of 57 killings per 100,000 people, the city's worst homicide rate on record."
It had almost nowhere to go but down.
Fact Check False … those are mostly peaceful confrontations with brief periods of violence.
Dang! Philly can’t even win in this league. Well they are doing a tad better than the Eagles.
There was peace when white millenials were growing up, blacks and mexicans became like them look at the millenial nba players like steph curry ,lebron ,harden nice every one of them ,white gen z are bat shit crazy , blacks and latinos are as usual copying whites, be ready for more violence as more of this age gruop become teens and turn twenty.
Psychoactive drugs, CRT indoctrination, and invasive tech did a number on those lil' SOBs.
Meanwhile:
Maryland county to pay $20mn in ‘heinous,’ ‘brutal’ & ‘senseless’ shooting of HANDCUFFED black man by black police officerFollowed by heartfelt encomiums. There must be more to this story ... guess a black guy pissed off a black officer - open to more spontaneous responses - enough.Replies: @anon, @Cloudbuster, @TomSchmidt, @DCThrowback
And who pays? Every time? Taxpayers.
It would be interesting to know if the quality of victims has changed or if it is just an increase of b on b killings. Are we seeing an increase of hate crimes or is this just a case of the usual suspects being more agitated than normal?
I wonder why murders are down slightly, 2020 cf. 2019.Replies: @stillCARealist, @Mark Roulo, @Tony, @Ga On My Mind
All the people who needed to be murdered were taken out in the last few years. Some guys just have it comin’, and everybody knows it. A crime wave is a good time to wreck some vengeance.
Monday Night news..
‘Yo ass goin’ viral’: Antifa, BLM protesters surround Portland police car, play victim when officer draws weapon
(((Tactics))):
White Helmet tactics:
(It’s an antique Colt 1911, too. Shit’s more than 100 years old.)
More White Helmet tactics:
Can anyone explain this? I guess it’s all about bypassing the local DA?
I’m pretty sure the political alignment (often nominal) of the people committing the murders shows a stark party difference.
I am amazed that Baltimore and Newark saw reductions in murder. Who saw that coming?
Meanwhile:
Maryland county to pay $20mn in ‘heinous,’ ‘brutal’ & ‘senseless’ shooting of HANDCUFFED black man by black police officerFollowed by heartfelt encomiums. There must be more to this story ... guess a black guy pissed off a black officer - open to more spontaneous responses - enough.Replies: @anon, @Cloudbuster, @TomSchmidt, @DCThrowback
There must be more to this story … guess a black guy pissed off a black officer
No body cam, no clear witnesses, and the officer isn’t talking to the press (which is smart). But that won’t prevent the press from trying him in absentia.
The criminal trial hasn’t occurred yet. It will be interesting to see what comes out then.
Meanwhile:
Maryland county to pay $20mn in ‘heinous,’ ‘brutal’ & ‘senseless’ shooting of HANDCUFFED black man by black police officerFollowed by heartfelt encomiums. There must be more to this story ... guess a black guy pissed off a black officer - open to more spontaneous responses - enough.Replies: @anon, @Cloudbuster, @TomSchmidt, @DCThrowback
One thing I learned from that case: if you’re going to carry a gun, make sure that you have insurance in case you need to use it. It’s about $500 a year, from what I looked up, and it seems to be well worth it. I’m sure the thought of having to come up with the money to pay for a lawyer, and having none, finally drove him over the edge.
https://www.uslawshield.com/about-us-2/Replies: @International Jew
Maybe Indians or Aleuts? That far north is not healthy for darker-skinned people unless they get extra Vitamin D.
Meanwhile:
Maryland county to pay $20mn in ‘heinous,’ ‘brutal’ & ‘senseless’ shooting of HANDCUFFED black man by black police officerFollowed by heartfelt encomiums. There must be more to this story ... guess a black guy pissed off a black officer - open to more spontaneous responses - enough.Replies: @anon, @Cloudbuster, @TomSchmidt, @DCThrowback
The cop was an AA PoS hire with a record of being a PoS.
Check out who runs PG county for a better explanation.
I wonder what’s up with Lubbock? Mostly White and Hispanic, and home of Texas Tech University. Increasing from 9 to 22 calls for some sort of explanation.
Regarding the recent uptick in homicides:
“Homicides are specifically a situation in which one person takes the life of another. It can include and does include murders. But, it’s not necessarily only murders, it can also be things like home invasions where the original victim of that home invasion then shot and killed that person coming into their home. It can also include officer involved shootings...and there are a couple of officer involved shootings in that time-period for the year and one of those home invasion situations,” said LPD Public Information Officer, Allison Matherly.
Matherly says many of these homicides show a distinct pattern of illegal activity before turning deadly, “The vast majority of the cases that we do have as homicides are really not random situations. They’re situations where people are already engaging in high-risk activities such as dealing drugs. Or, again, home invasions, things like that.”
While Lubbock is currently averaging a little more than four homicides a month, Matherly says the Police Department wants residents to remember something specific, "The average Lubbock citizen who’s going about their life, not engaging in high-risk behaviors is really not going to find themselves in a situation that is concerning. We’re not seeing random crime across town that’s leading to homicides.”
Coronachan has just made 7 digits!
https://www.epsilontheory.com/wp-content/uploads/epsilon-theory-one-million-dollars-september-15-2015-austin-powers.jpg
One million deaths!
On an Earth where 57 million die each year and 141 million are born each year.
Covid-19 pandemic has cost the world's economy $3.8TRILLION 'and made 147 million people unemployed', study claimsReplies: @Mr. Anon
A related story from 1918: lone Serbian gunman costs the World $340 billion, kills 14 million people.
I wonder why murders are down slightly, 2020 cf. 2019.Replies: @stillCARealist, @Mark Roulo, @Tony, @Ga On My Mind
Baltimore’s slight drop from 2019 to 2020 is well within random statistical variation year-to-year (I’ve been tracking this since Freddie Grey and have monthly data back to 2007). I haven’t updated my tables for 2020 data, but the basic conclusion is clear.
You can see the data here: http://mistybeach.com/mark/BaltimoreHomicides.html
This article was just published about Memphis. They are having a bad year.
https://www.commercialappeal.com/story/news/2020/09/29/memphis-passes-homicides-record-2016-three-months-go-2020/3570806001/
What is interesting is that this page has gone missing (I see it in a DuckDuckGo search).
https://www.memphistn.gov/government/police-department/crime_statistics
But the actual data page it used to link to is still there.
https://data.memphistn.gov/
If you go to the public safety page, zoom way out to get a single number, and select the four types of homicide it is easy to generate totals for different date ranges.
1/1/19 – 9/29/19 149 homicides
1/1/20 – 9/29/20 221 homicides
That 48% increase looks like it would be in the top third of your table.
On the other hand, Pittsburgh has a pretty good crime data page.
https://pittsburghpa.gov/publicsafety/crime-data
but it has only been updated through 7/31/20. (for the overall crime data page, the homicides page has only been updated to 6/14/20 when they show 17 vs. 38 in 2019, must be using slightly different definition than “criminal homicide” I looked at below)
They show 23 criminal homicides year to date vs. 35 for all of 2019. I don’t see any way to look at finer grained date ranges rather than years.
Surprising to see how many Texas cities are above the national average. Lubbock, Forth Worth, Arlington, Austin, and Houston are all at least 30% and above. San Antonio (26%) is slightly below the national average, and only Dallas (4%) and Plano (0%) are well below it.
What’s also strange is that Dallas, which is probably Texas’s blackest large city, hasn’t seen much of an increase at all, but nearby Forth Worth and Arlington are around double the national increase with proportionally smaller black populations.
Are Hispanics behaving badly without the police boot on their necks like blacks are. Are cops in LA or El Paso going to doughnut shops instead of policing Hispanic areas?
I can imagine that illegal immigration and asylum fraud immigration is down because of SARS-CoV-2, but there are so many Hispanics squatting here already that I doubt a few months of illegals makes much difference. Don’t get me wrong, every little bit helps!
Are Hispanics treating it like Minority Lives Matter? I can totally imagine that the progressives who do the thinking for Hispanics are working on LxLM, Latinx Lives Matter*,but I think it will be focused on cranking up the infestation, along with amnesty and prizes for illegals who have been squatting for a while. But I could be wrong, and maybe the underclass Latinkques (plural of Latinx) who assimilated to black culture would prefer legal immunity to more competition.
Imagine the media uproar over the name! They would get so much publicity from a media who could not help covering it and a progressive establishment that was very torn. I mean, you can’t say all lives matter, because that does not give proper primacy to blacks, but mestizos and mulattos are also sacred minorities. Not saying Latinx Lives Matter is insensitivo, but immigration restrictionists won’t be saying it, and they don’t want to sound like patriots. It will be like a computer on Star Trek when presented with a logical paradox. They’ll repeat a few times, and then their butts catch fire. Why their butts? What do you think progressives have their head up, hmm?
This can be confusing since Dallas is often lumped with Ft. Worth and the entire Metroplex which in total is larger than Houston. Dallas "metro" population is 2.636 million. Houston metro area (including Galveston & close suburbs) is 7.07 million (both 2019).
The non Dallas/Houston suburbs and nearby cities tend to be less black than the actual major cities themselves, though Galveston may be an exception.Replies: @Pincher Martin, @res
You’ve noticed that too. It started sometime in the last 10 years or so. Liberals now seem to place more trust in the CIA and the FBI than they do in any other organizations, perhaps even more than they do in the Press. Perhaps it’s down to all the movies and television they’ve seen – movies and TV that increasingly portray such organizations as being righteous and super-competent. Guess who offers technical advice and support for such programs?
Today, the CIA is portrayed in TV shows like Jack Ryan. Forty years ago it was portrayed in movies like Three Days of the Condor.
Thirty years ago, liberals used to have “Question Authority” bumper-stickers on their cars. I haven’t seen to many of them recently.
Try mapping murder by congressional district. Nebraska is an easy study, with only three districts. Since Jimmy Carter carried Texas in 1976, Democrats have won precisely one elector in that middle strip of states reaching from Texas to North Dakota. That was from Nebraska’s Second District, in 2008. Omaha’s, where what diversity in Nebraska is concentrated.
Gun Violence Archive gives stats for every district. With a third of the state’s population, NE-2 hosted 58% of the gun homicides for the perion shown and, more telling, 79% of the injuries. Omahans are shooting and missing!
Mississippi is similar, if less stark. About 40% of homicides and injuries take place in one of the four districts, twice the rate of the others. Alabama’s Seventh has 38% of the state’s murders and 40% of the injuries. These are the only districts in those states represented by a Democrat. Causation is a chicken-and-egg affair.
https://www.gunviolencearchive.org/congress/ne/02
https://www.gunviolencearchive.org/congress/ms/02
https://www.gunviolencearchive.org/congress/al/07
WaPo has this map of crime by county. Joe’s Delaware sticks out in the relatively placid Northeast, as does lesbian central Massachusetts. Appalachia is less violent than the flatter parts of the states it’s in. Kentucky is remarkably safe; must be the horses.
I believe this is the source of the data for the WaPo map. Might be useful.
Uniform Crime Reporting Program Data: County-Level Detailed Arrest and Offense Data, United States, 2014 (ICPSR 36399)
https://www.icpsr.umich.edu/web/NACJD/studies/36399
Here is a list of UCR data there.
https://www.icpsr.umich.edu/web/pages/NACJD/guides/ucr.html
I see 2009-2014 and 2016 county level arrest data there. It is very interesting that I can't find the 2015 data there. I do find the 2015 data at
https://www.openicpsr.org/openicpsr/project/108164/version/V3/view
Note that the data has breakdowns by race, sex, and age.
What is the relevance?
The last column, T, is the number of standard deviations:
T = (m2020-m2019)/sqrt((m2020+m2019)/2)
City m2019 m2020 T
Chicago 374 560 8.61
Milwaukee 66 134 6.80
New_York 236 327 5.42
Louisville 62 104 4.61
New_Orleans 87 133 4.39
Philadelphia 243 315 4.31
St_Louis 136 185 3.87
Minneapolis 32 57 3.75
Houston 176 229 3.72
Lubbock 9 22 3.30
Atlanta 69 97 3.07
Detroit 187 230 2.98
Indianapolis 98 129 2.91
Fort_Worth 31 49 2.85
Denver 43 63 2.75
Kansas_City 113 144 2.73
Toledo 22 36 2.60
Cleveland 88 114 2.59
Fort_Wayne 13 24 2.56
Boston 30 45 2.45
Nashville 54 73 2.38
Cincinnati 54 73 2.38
Miami 60 77 2.05
Lexington 16 25 1.99
San_Antonio 66 83 1.97
Colorado_Springs 14 22 1.89
Austin 23 33 1.89
Aurora 15 23 1.84
San_Bernardino 25 35 1.83
Oakland 52 66 1.82
Los_Angeles 197 223 1.79
Greensboro 30 40 1.69
Riverside 7 12 1.62
Tulsa 37 47 1.54
Jacksonville 93 108 1.50
Arlington 13 19 1.50
Charlotte 73 84 1.24
Washington 126 140 1.21
Seattle 23 29 1.18
San_Diego 24 30 1.15
Sacramento 18 23 1.10
San_Francisco 23 27 0.80
Portland 22 25 0.62
Stockton 28 31 0.55
Dallas 151 157 0.48
Bakersfield 27 29 0.38
Long_Beach 18 19 0.23
Omaha 23 24 0.21
Plano 1 1 0.00
Lincoln 4 4 0.00
Las_Vegas 64 64 0.00
Baltimore 247 240 -0.45
San_Jose 24 21 -0.63
Durham 22 18 -0.89
Newark 37 31 -1.03
Virginia_Beach 9 6 -1.10
Chula_Vista 11 7 -1.33
Oklahoma_City 39 31 -1.35
Anchorage 20 6 -3.88Replies: @International Jew, @res, @silviosilver
Thanks! That is a very interesting way of looking at this. I like that the result is expressed in SDs which I think makes it more interpretable in terms of how uncommon a change of that magnitude is.
Your comment made me curious, so I looked around and found this 2017 paper.
Monitoring Volatile Homicide Trends Across U.S. Cities
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2977556
Note their method and conclusion.
Any thoughts on using the Poisson vs. binomial distribution for modeling homicides? This is getting beyond my statistical pay grade, but it looks like the difference is small (in the limit they are the same) and if anything using Poisson understates the rarity of evens (mean = variance vs. mean > variance for binomial).
https://keydifferences.com/difference-between-binomial-and-poisson-distribution.html
They include a link to data and code for replication, but use SPSS which complicates things a bit for those who don’t have it. This R package might be helpful.
https://www.r-bloggers.com/2013/11/spss-to-r-an-r-package-to-convert-spss-syntax/
I’m trying to work out how to interpret things given that they draw very different conclusions about the presence of a Ferguson effect based on what look like fairly similar numbers. Taking the Baltimore panel of Figure 4 as an example we see the 2014 homicide rate was about 33 compared to about 55 for 2015.
If I understand your method correctly this would indicate a (55 – 33) / sqrt((55+33)/2) = 3.3 SD change. That is about a 1 in 2000 event. Yet looking at their graphic we see 2015 barely falls outside their 50% prediction interval and is well inside their 80% prediction interval.
Those approaches give vastly different indicators of how rare an event the Baltimore 2014-2015 change in homicides was. Perhaps a useful check would be to apply your methodology to each yearly change and see how rare all of those look?
What do you (or anyone else who is statistically literate enough to understand all of this) think?
"If I understand your method correctly this would indicate a (55 – 33) / sqrt((55+33)/2) = 3.3 SD change. That is about a 1 in 2000 event. Yet looking at their graphic we see 2015 barely falls outside their 50% prediction interval and is well inside their 80% prediction interval."
You shouldn't be looking at rates isolated from the population size (as random fluctuations in the rate will be smaller in large populations). According to wikipedia the actual numbers were 211 homicides in 2014 and 344 homicides in 2015 (for rates of 33.8 and 55.4 per 100,000). We can then ask what is the chance of flipping a coin 555 (211+344) times and getting 211 heads or less. The answer is less than 1 chance in 100,000,000 so probably not a random event ( (this is a one-sided test, you can multiply by 2 if you prefer a 2-sided test, still doesn't look random). This assumes the population didn't change but if anything it appears from the calculated rates per 100,000 that the population went down a little so cannot explain a raise in homicides.
If you look at several years the effect is even more obvious. Again according to wikipedia, the number of homicides for years 2010-2019 were 223, 196, 218, 233, 211, 344, 318, 343, 309, 348 respectively.Replies: @Steve Sailer, @res
What the hell is going on in Lubbock, TX?
“Murders Are Rising. Blaming a Party Doesn’t Add Up.”
yes, don’t blame the party that is causing it. it’s definitely NOT democrats and democrat politicians who want to attack police officers, defund the police, reduce police presence, and stop officers from reducing crime.
https://www.gunviolencearchive.org/congress/ne/02
https://www.gunviolencearchive.org/congress/ms/02
https://www.gunviolencearchive.org/congress/al/07
WaPo has this map of crime by county. Joe's Delaware sticks out in the relatively placid Northeast, as does lesbian central Massachusetts. Appalachia is less violent than the flatter parts of the states it's in. Kentucky is remarkably safe; must be the horses.
https://www.washingtonpost.com/graphics/national/crime-rates-by-county/img/violent-980.jpg?c=367Replies: @res
Thanks. That looks like a useful source of data. Do they offer any way to do bulk data export? It seems like their data would be most useful if it could be recast into a per capita form and linked with other data (for example, Congressional party, as you did).
I believe this is the source of the data for the WaPo map. Might be useful.
Uniform Crime Reporting Program Data: County-Level Detailed Arrest and Offense Data, United States, 2014 (ICPSR 36399)
https://www.icpsr.umich.edu/web/NACJD/studies/36399
Here is a list of UCR data there.
https://www.icpsr.umich.edu/web/pages/NACJD/guides/ucr.html
I see 2009-2014 and 2016 county level arrest data there. It is very interesting that I can’t find the 2015 data there. I do find the 2015 data at
https://www.openicpsr.org/openicpsr/project/108164/version/V3/view
Note that the data has breakdowns by race, sex, and age.
Slightly less awful. Not a whole lot less awful.
A better analysis would be what is the variation over the last ten years (correct for population size change) and see how far outside the anticipated homicides are cities. Steve has done a very good analysis of showing the increase in Baltimore but is plus or minus 30% that strange when it comes to rare events such as homicides.
One probably should assume that cities with low homicide rates are not represented by a normal distribution but maybe a Poisson distribution.
See a dentist about that tooth. Forthwith.
Not sure how that happened.
Prof., if one of our resident statisticans wishes to do some onerous research it would be interesting to know the number of quality victims, such as black women and children, killed during this muderous spree.
Your comment made me curious, so I looked around and found this 2017 paper.
Monitoring Volatile Homicide Trends Across U.S. Cities
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2977556Note their method and conclusion.Any thoughts on using the Poisson vs. binomial distribution for modeling homicides? This is getting beyond my statistical pay grade, but it looks like the difference is small (in the limit they are the same) and if anything using Poisson understates the rarity of evens (mean = variance vs. mean > variance for binomial).
https://keydifferences.com/difference-between-binomial-and-poisson-distribution.html
They include a link to data and code for replication, but use SPSS which complicates things a bit for those who don't have it. This R package might be helpful.
https://www.r-bloggers.com/2013/11/spss-to-r-an-r-package-to-convert-spss-syntax/
I'm trying to work out how to interpret things given that they draw very different conclusions about the presence of a Ferguson effect based on what look like fairly similar numbers. Taking the Baltimore panel of Figure 4 as an example we see the 2014 homicide rate was about 33 compared to about 55 for 2015.
If I understand your method correctly this would indicate a (55 - 33) / sqrt((55+33)/2) = 3.3 SD change. That is about a 1 in 2000 event. Yet looking at their graphic we see 2015 barely falls outside their 50% prediction interval and is well inside their 80% prediction interval.
Those approaches give vastly different indicators of how rare an event the Baltimore 2014-2015 change in homicides was. Perhaps a useful check would be to apply your methodology to each yearly change and see how rare all of those look?
What do you (or anyone else who is statistically literate enough to understand all of this) think?Replies: @International Jew, @James B. Shearer
I think we talked about this guy here on isteve, a few years ago. What he does to make the Ferguson effect disappear (in his mind anyway) is to say that the increase of the last few years has been small compared to the decline since the peak crime year of 1991.
In (slightly more) technical terms, he’s pretending that the murder numbers of the last 60 years have been random draws out of a hat, and that the increase in recent years is small compared to the standard deviation of all the years since 1960.
Which would be a sensible way to go if we didn’t know anything about the causes of crime, ie we knew nothing of the 1960s’ experiment with leniency, nothing about Crack, nothing about the Giuliani-Bratton-led get-tough approach of the 90s, nothing about the late Obama war on police. He also likes to look at the country as a whole, as though we know nothing about the circumstances of individual cities.
It looks like a willful whitewash to me. But after another few years like this one he’ll have run out of rope anyway because we’ll be back to 1991 levels of crime.
What we are trying to say (IMO) is that the Ferguson effect is clearly not random chance. And I think your methodology is appropriate for that. Though perhaps it would make sense to use an average of N prior years to calculate the prevailing mean and thus the random variance/SD? I think including the post-spike data tends to inflate the variance. But is that a good thing because it makes the estimates more conservative?
Mrs. FPD72 and I use U.S. Lawshield. The total cost for enrollment and annual premium for two, with multi-state coverage, was $350. It was promoted by the instructor in our License to Carry (LTC) class. I think we got a discount due to the class. It provides full attorney counsel and representation for both criminal charges and civil lawsuits for any actions taken in self-defense. If you’re involved in an incident, after calling 911, you call them and then they either provide counsel over the phone or actually send an attorney to the scene. They also provide training and education through live seminars and on the net, free to all clients.
https://www.uslawshield.com/about-us-2/
LOL!
Not sure how that happened.
Stick, I am not trying to be argumentative, but a 3% drop in Baltimore is hardly cause to celebrate. not that you are. We’ve become so used to murder in certain communities that a few less bodies is note worthy.
Lubbock’s population is 9% black (mostly on the east side) and 32% Hispanic (mostly on the north side). A Hispanic biker gang, the Diablos, is very active in drug trafficking, oftentimes with fatal results.
Regarding the recent uptick in homicides:
“Homicides are specifically a situation in which one person takes the life of another. It can include and does include murders. But, it’s not necessarily only murders, it can also be things like home invasions where the original victim of that home invasion then shot and killed that person coming into their home. It can also include officer involved shootings…and there are a couple of officer involved shootings in that time-period for the year and one of those home invasion situations,” said LPD Public Information Officer, Allison Matherly.
Matherly says many of these homicides show a distinct pattern of illegal activity before turning deadly, “The vast majority of the cases that we do have as homicides are really not random situations. They’re situations where people are already engaging in high-risk activities such as dealing drugs. Or, again, home invasions, things like that.”
While Lubbock is currently averaging a little more than four homicides a month, Matherly says the Police Department wants residents to remember something specific, “The average Lubbock citizen who’s going about their life, not engaging in high-risk behaviors is really not going to find themselves in a situation that is concerning. We’re not seeing random crime across town that’s leading to homicides.”
It is, but that is where they focus on data analysis.
https://www.nytimes.com/newsletters/upshot
Buffalo, its always interesting when a heavy hitter starts batting .180. My guess is Baltimore and Newark are running out of targets.
Commenter Mark Roulo (presently #40 supra) linked a page where he looked month-by-month at Baltimore’s murder tally. Near the bottom, the graph “‘Excess’ Baltimore Homicides Per Month Compared to 2007-2014 Monthly Average” tells a tale that is much starker than the one told by academics Wheeler and Kovandzic in their Fig. 4, linked by res (currently #46). (That paper dates from 2017, and has only the second half of 2015 as post-Freddy Gray murder data.)
That would make sense. I did a quick search for that paper on iSteve before positing my earlier comment, but did not see it. I just searched again for Andrew Wheeler and did not see that either though. We may just have lumped him in with all of the other similar arguments happening at the time.
Thanks. That was my sense as well, but I wasn’t feeling confident about it. I guess the key question is how well the prediction intervals reflect typical random change. The problem being that there are all sorts of non-random trends and events contributing to the actual data.
Agreed about the first sentence. The second is more complicated. In 2015 Baltimore already exceeded their highs from the early 1990’s, while Dallas, NYC, and Chicago were a fifth, sixth, and third of their 1991 rates respectively. That is a long way to go to catch up.
I think that highlights the way to think about this. We are not trying to say the Ferguson effect is unprecedented historically. It would be more than bad enough to be comparable to one of the events you list.
What we are trying to say (IMO) is that the Ferguson effect is clearly not random chance. And I think your methodology is appropriate for that. Though perhaps it would make sense to use an average of N prior years to calculate the prevailing mean and thus the random variance/SD? I think including the post-spike data tends to inflate the variance. But is that a good thing because it makes the estimates more conservative?
Thanks for that link. I thought the Baltimore panel of Figure 4 was quite stark (the jump from 32 to 55). What your link adds is that the number of homicides never really came down after 2015. One might argue that showing the rates as deviations from the average overemphasizes the changes. This is similar to how omitting part of the y-axis overemphasizes changes in graphs (a common tactic and complaint). Except in this case we are shifting the axis all the way to the average value.
One other point worth noting from your link. He gives the 2007-2014 mean and SD as 222.6 and 15.3. Using International Jew’s Poisson approach we would calculate a SD of sqrt(222.6) = 14.9 which is pretty close to that 15.3. I suspect the comparison would be even closer if we eliminated 2007 which seems like an obvious outlier over that interval.
If you have red blood in your veins, this'll make it boil.
https://www.takimag.com/article/innocent-until-proven-trump-supporter/Replies: @Charon, @El Dato, @SunBakedSuburb, @Glt, @Ben tillman
“this’ll make it boil”
Less boiling more leisure. The world is making it difficult to relax. Weed and headphones; just like PB&J.
“white gen z are bat shit crazy”
Psychoactive drugs, CRT indoctrination, and invasive tech did a number on those lil’ SOBs.
That inspires a question for the BLM era. How have Hispanics responded to BLM. Not that that long ago, cops in the Northeast were hard on (mulatto) hispanics. I’ve heard cops in California don’t arrest illegals for minor crimes, but I haven’t heard anything about the Southwest. Mr Unz aside, most hbders believe Hispanics are crime-prone, though less so than blacks.
Are Hispanics behaving badly without the police boot on their necks like blacks are. Are cops in LA or El Paso going to doughnut shops instead of policing Hispanic areas?
I can imagine that illegal immigration and asylum fraud immigration is down because of SARS-CoV-2, but there are so many Hispanics squatting here already that I doubt a few months of illegals makes much difference. Don’t get me wrong, every little bit helps!
Are Hispanics treating it like Minority Lives Matter? I can totally imagine that the progressives who do the thinking for Hispanics are working on LxLM, Latinx Lives Matter*,but I think it will be focused on cranking up the infestation, along with amnesty and prizes for illegals who have been squatting for a while. But I could be wrong, and maybe the underclass Latinkques (plural of Latinx) who assimilated to black culture would prefer legal immunity to more competition.
Imagine the media uproar over the name! They would get so much publicity from a media who could not help covering it and a progressive establishment that was very torn. I mean, you can’t say all lives matter, because that does not give proper primacy to blacks, but mestizos and mulattos are also sacred minorities. Not saying Latinx Lives Matter is insensitivo, but immigration restrictionists won’t be saying it, and they don’t want to sound like patriots. It will be like a computer on Star Trek when presented with a logical paradox. They’ll repeat a few times, and then their butts catch fire. Why their butts? What do you think progressives have their head up, hmm?
How is this even remotely possible? Are they not including any of the damage from the riots? Is it not being reported for some reason? This seems like a situation similar to the economists who say with a straight face that the mass Cuban immigration into Miami had no effect, while ignoring the massive organized crime wave it spurred.
Statistics: torturing data until it confesses
I wonder why murders are down slightly, 2020 cf. 2019.Replies: @stillCARealist, @Mark Roulo, @Tony, @Ga On My Mind
Newark also is surprisingly an outlier given the negro situation there.
Joe has clearly had some major plastic surgery done recently. Is that why they were keeping him under wraps for so long? Because he had bandages/scars?
He looks like the aliens from V. Almost human but not quite.
The last column, T, is the number of standard deviations:
T = (m2020-m2019)/sqrt((m2020+m2019)/2)
City m2019 m2020 T
Chicago 374 560 8.61
Milwaukee 66 134 6.80
New_York 236 327 5.42
Louisville 62 104 4.61
New_Orleans 87 133 4.39
Philadelphia 243 315 4.31
St_Louis 136 185 3.87
Minneapolis 32 57 3.75
Houston 176 229 3.72
Lubbock 9 22 3.30
Atlanta 69 97 3.07
Detroit 187 230 2.98
Indianapolis 98 129 2.91
Fort_Worth 31 49 2.85
Denver 43 63 2.75
Kansas_City 113 144 2.73
Toledo 22 36 2.60
Cleveland 88 114 2.59
Fort_Wayne 13 24 2.56
Boston 30 45 2.45
Nashville 54 73 2.38
Cincinnati 54 73 2.38
Miami 60 77 2.05
Lexington 16 25 1.99
San_Antonio 66 83 1.97
Colorado_Springs 14 22 1.89
Austin 23 33 1.89
Aurora 15 23 1.84
San_Bernardino 25 35 1.83
Oakland 52 66 1.82
Los_Angeles 197 223 1.79
Greensboro 30 40 1.69
Riverside 7 12 1.62
Tulsa 37 47 1.54
Jacksonville 93 108 1.50
Arlington 13 19 1.50
Charlotte 73 84 1.24
Washington 126 140 1.21
Seattle 23 29 1.18
San_Diego 24 30 1.15
Sacramento 18 23 1.10
San_Francisco 23 27 0.80
Portland 22 25 0.62
Stockton 28 31 0.55
Dallas 151 157 0.48
Bakersfield 27 29 0.38
Long_Beach 18 19 0.23
Omaha 23 24 0.21
Plano 1 1 0.00
Lincoln 4 4 0.00
Las_Vegas 64 64 0.00
Baltimore 247 240 -0.45
San_Jose 24 21 -0.63
Durham 22 18 -0.89
Newark 37 31 -1.03
Virginia_Beach 9 6 -1.10
Chula_Vista 11 7 -1.33
Oklahoma_City 39 31 -1.35
Anchorage 20 6 -3.88Replies: @International Jew, @res, @silviosilver
It would have meant much more work in tracking down the relevant data (so I don’t blame you for not undertaking it), but it would have been better to use a longer-term mean – say, the last five years.
Also, using 2020 data to calculate the mean has the effect of understating the variance. Since what we’re really interested in is assessing is just how big of an anomaly 2020 has been it I think it would have been better to just use 2019 as the mean.
It’s hard to detect any rhyme or reason in the fluctuations. Sailer claims to see some effect in the extent of BLM’s influence in a city, but it doesn’t seem to amount to very much really, not when you consider the range of cities on that list.
Newark no es tan “negro” como crees.
Your comment made me curious, so I looked around and found this 2017 paper.
Monitoring Volatile Homicide Trends Across U.S. Cities
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2977556Note their method and conclusion.Any thoughts on using the Poisson vs. binomial distribution for modeling homicides? This is getting beyond my statistical pay grade, but it looks like the difference is small (in the limit they are the same) and if anything using Poisson understates the rarity of evens (mean = variance vs. mean > variance for binomial).
https://keydifferences.com/difference-between-binomial-and-poisson-distribution.html
They include a link to data and code for replication, but use SPSS which complicates things a bit for those who don't have it. This R package might be helpful.
https://www.r-bloggers.com/2013/11/spss-to-r-an-r-package-to-convert-spss-syntax/
I'm trying to work out how to interpret things given that they draw very different conclusions about the presence of a Ferguson effect based on what look like fairly similar numbers. Taking the Baltimore panel of Figure 4 as an example we see the 2014 homicide rate was about 33 compared to about 55 for 2015.
If I understand your method correctly this would indicate a (55 - 33) / sqrt((55+33)/2) = 3.3 SD change. That is about a 1 in 2000 event. Yet looking at their graphic we see 2015 barely falls outside their 50% prediction interval and is well inside their 80% prediction interval.
Those approaches give vastly different indicators of how rare an event the Baltimore 2014-2015 change in homicides was. Perhaps a useful check would be to apply your methodology to each yearly change and see how rare all of those look?
What do you (or anyone else who is statistically literate enough to understand all of this) think?Replies: @International Jew, @James B. Shearer
“I’m trying to work out how to interpret things given that they draw very different conclusions about the presence of a Ferguson effect based on what look like fairly similar numbers. Taking the Baltimore panel of Figure 4 as an example we see the 2014 homicide rate was about 33 compared to about 55 for 2015.”
“If I understand your method correctly this would indicate a (55 – 33) / sqrt((55+33)/2) = 3.3 SD change. That is about a 1 in 2000 event. Yet looking at their graphic we see 2015 barely falls outside their 50% prediction interval and is well inside their 80% prediction interval.”
You shouldn’t be looking at rates isolated from the population size (as random fluctuations in the rate will be smaller in large populations). According to wikipedia the actual numbers were 211 homicides in 2014 and 344 homicides in 2015 (for rates of 33.8 and 55.4 per 100,000). We can then ask what is the chance of flipping a coin 555 (211+344) times and getting 211 heads or less. The answer is less than 1 chance in 100,000,000 so probably not a random event ( (this is a one-sided test, you can multiply by 2 if you prefer a 2-sided test, still doesn’t look random). This assumes the population didn’t change but if anything it appears from the calculated rates per 100,000 that the population went down a little so cannot explain a raise in homicides.
If you look at several years the effect is even more obvious. Again according to wikipedia, the number of homicides for years 2010-2019 were 223, 196, 218, 233, 211, 344, 318, 343, 309, 348 respectively.
https://stats.stackexchange.com/questions/15371/how-to-calculate-a-confidence-level-for-a-poisson-distributionCould you elaborate on this method? Is it commonly used? The 1 chance in 1e8 seemed improbably small to me, but I did a quick check in R and got an even smaller number (2e-17) assuming a mean of 211 and an observation of 344.
ppois(344, lambda=211, lower=FALSE)
Using the 5 year mean and 2015 numbers from ic1000's link I saw:
ppois(342, lambda=222.6, lower=FALSE) = 5e-14
I think it would be better to use International Jew's method with a 5 year prior mean and current observation (rather than computing the mean of the prior and current year), but I think just using the prior year and current year would be good enough.
That would give (55.4 – 33.8) / sqrt((33) = 3.8 SD change. So a bit larger than the earlier 3.3 SD estimate, but not that much.Agreed. I picked Baltimore because it seems so blatantly nonrandom, yet the Wheeler paper seemed to question even it. Can you comment on the analysis in that paper?
Could you elaborate on how you would change International Jew's comment 18 to reflect both individual city and overall probability the change was nonrandom? I think his method his good for assessing the magnitude of the change (do you agree?), but the map to chance the change was random is imperfect because of your sample size point.
Thanks for your response!
P.S. Regarding rates vs. absolute counts, I'm not sure whether the more important effect to address is changing population size or small sample noise. I think the best approach would be to address both, but that makes things more difficult. How would you recommend doing that?Replies: @James B. Shearer, @James B. Shearer, @James B. Shearer
"If I understand your method correctly this would indicate a (55 – 33) / sqrt((55+33)/2) = 3.3 SD change. That is about a 1 in 2000 event. Yet looking at their graphic we see 2015 barely falls outside their 50% prediction interval and is well inside their 80% prediction interval."
You shouldn't be looking at rates isolated from the population size (as random fluctuations in the rate will be smaller in large populations). According to wikipedia the actual numbers were 211 homicides in 2014 and 344 homicides in 2015 (for rates of 33.8 and 55.4 per 100,000). We can then ask what is the chance of flipping a coin 555 (211+344) times and getting 211 heads or less. The answer is less than 1 chance in 100,000,000 so probably not a random event ( (this is a one-sided test, you can multiply by 2 if you prefer a 2-sided test, still doesn't look random). This assumes the population didn't change but if anything it appears from the calculated rates per 100,000 that the population went down a little so cannot explain a raise in homicides.
If you look at several years the effect is even more obvious. Again according to wikipedia, the number of homicides for years 2010-2019 were 223, 196, 218, 233, 211, 344, 318, 343, 309, 348 respectively.Replies: @Steve Sailer, @res
The big, long-lasting increase in Baltimore homicides can be dated pretty precisely to the BLM Freddie Gray riot of April 25, 2015.
http://mistybeach.com/mark/BaltimoreHomicides.html
This viewer for Baltimore homicides might be of interest.
https://homicides.news.baltimoresun.com/
It includes 2007-2020, but unfortunately does not allow looking at pre-2020 data with a granularity other than years.
I was wondering why 2007 seemed like an outlier. This looks like a good candidate for the reason.
http://foplodge4.org/baltimore-falls-out-of-top-5-murder-rate-cities/
Agreed. The monthly stats at the link ic1000 gave show that quite clearly.
http://mistybeach.com/mark/BaltimoreHomicides.html
This viewer for Baltimore homicides might be of interest.
https://homicides.news.baltimoresun.com/
It includes 2007-2020, but unfortunately does not allow looking at pre-2020 data with a granularity other than years.
I was wondering why 2007 seemed like an outlier. This looks like a good candidate for the reason.
http://foplodge4.org/baltimore-falls-out-of-top-5-murder-rate-cities/
Interesting that you think of Plano Illinois and not the 300,000 people of Plano Texas.
https://www.uslawshield.com/about-us-2/Replies: @International Jew
You think you’d be able to count on them if you got yourself into a high-profile case? You don’t think they’d find a way to drop you, as soon as the mob marched to the home of their CEO? You think the lawyer they assigned you wouldn’t scram once the mob marched to his home?
How would the mob even know you have insurance? And dropping a client in a high profile case isn't good for business. Anyway most cases aren't high profile but you still might need good (and expensive) legal representation.
US CCA seems more solid than the NRA. My guess is they wouldn't blink, but who knows.Replies: @Goddard
“You think you’d be able to count on them if you got yourself into a high-profile case? You don’t think they’d find a way to drop you, as soon as the mob marched to the home of their CEO? …”
How would the mob even know you have insurance? And dropping a client in a high profile case isn’t good for business. Anyway most cases aren’t high profile but you still might need good (and expensive) legal representation.
Do you advise going without any coverage? Or is there an organization you can recommend that will not wilt?
US CCA seems more solid than the NRA. My guess is they wouldn’t blink, but who knows.
OT
Edward Hopper cancellation attempt by the New York Times:
https://althouse.blogspot.com/2020/09/when-great-painter-edward-hopper-was.html
The gist: He copied other paintings as a teenager while learning to paint, and this means that the woke should cancel this white man, through a tortuous train of logic. Copying paintings was entirely normal, and apparently when it was pointed out to the Times they added a sentence deep in the article, even though the entire article now becomes pointless.
The article is by a sibling of the New Yorker’s Adam “Even Hitler wasn’t Hitler before he was” Adam Gopnik.
Trivia: After he attained success Raymond Chandler wrote Erle Stanley Gardner a fan letter telling him that in learning to write he took one of Gardner’s novelettes, transcribed the bare plot, rewrote the novelette from that, compared it to the original, and did another rewrite. This is brilliant since the plot transcription taught him how plots are structured, and the rewrites taught him characterization, dialog, and point of view. I think he did something similar with James M. Cain, but superstitiously avoided it with Dashiell Hammett because he admired him so much.
If you have red blood in your veins, this'll make it boil.
https://www.takimag.com/article/innocent-until-proven-trump-supporter/Replies: @Charon, @El Dato, @SunBakedSuburb, @Glt, @Ben tillman
That probably should make it boil, but it just seems normal to me now. I guess I am blackpilled.
[Agree]
The now giant municipalities of the DFW Metroplex like Plano and Frisco are hard to get a sense of if you are outside Texas because they weren’t a big deal when one was a kid.
US CCA seems more solid than the NRA. My guess is they wouldn't blink, but who knows.Replies: @Goddard
I use USCCA.
"If I understand your method correctly this would indicate a (55 – 33) / sqrt((55+33)/2) = 3.3 SD change. That is about a 1 in 2000 event. Yet looking at their graphic we see 2015 barely falls outside their 50% prediction interval and is well inside their 80% prediction interval."
You shouldn't be looking at rates isolated from the population size (as random fluctuations in the rate will be smaller in large populations). According to wikipedia the actual numbers were 211 homicides in 2014 and 344 homicides in 2015 (for rates of 33.8 and 55.4 per 100,000). We can then ask what is the chance of flipping a coin 555 (211+344) times and getting 211 heads or less. The answer is less than 1 chance in 100,000,000 so probably not a random event ( (this is a one-sided test, you can multiply by 2 if you prefer a 2-sided test, still doesn't look random). This assumes the population didn't change but if anything it appears from the calculated rates per 100,000 that the population went down a little so cannot explain a raise in homicides.
If you look at several years the effect is even more obvious. Again according to wikipedia, the number of homicides for years 2010-2019 were 223, 196, 218, 233, 211, 344, 318, 343, 309, 348 respectively.Replies: @Steve Sailer, @res
Agreed. The question is how big a difference does that make. Calculating a confidence interval for a Poisson distribution is more complicated than I realized.
https://stats.stackexchange.com/questions/15371/how-to-calculate-a-confidence-level-for-a-poisson-distribution
Could you elaborate on this method? Is it commonly used? The 1 chance in 1e8 seemed improbably small to me, but I did a quick check in R and got an even smaller number (2e-17) assuming a mean of 211 and an observation of 344.
ppois(344, lambda=211, lower=FALSE)
Using the 5 year mean and 2015 numbers from ic1000’s link I saw:
ppois(342, lambda=222.6, lower=FALSE) = 5e-14
I think it would be better to use International Jew’s method with a 5 year prior mean and current observation (rather than computing the mean of the prior and current year), but I think just using the prior year and current year would be good enough.
That would give (55.4 – 33.8) / sqrt((33) = 3.8 SD change. So a bit larger than the earlier 3.3 SD estimate, but not that much.
Agreed. I picked Baltimore because it seems so blatantly nonrandom, yet the Wheeler paper seemed to question even it. Can you comment on the analysis in that paper?
Could you elaborate on how you would change International Jew’s comment 18 to reflect both individual city and overall probability the change was nonrandom? I think his method his good for assessing the magnitude of the change (do you agree?), but the map to chance the change was random is imperfect because of your sample size point.
Thanks for your response!
P.S. Regarding rates vs. absolute counts, I’m not sure whether the more important effect to address is changing population size or small sample noise. I think the best approach would be to address both, but that makes things more difficult. How would you recommend doing that?
You should be using 555/2 =277.5 as the mean.Replies: @res
It appears to me that the paper is not claiming that the change is nonrandom just that it is not dramatically larger than other nonrandom changes in Baltimore in the past.Replies: @res
It is easy to adjust my proposed test for a changing population size (as long as you know the sizes), just use a binomial distribution with p other than .5.Replies: @res
https://stats.stackexchange.com/questions/15371/how-to-calculate-a-confidence-level-for-a-poisson-distributionCould you elaborate on this method? Is it commonly used? The 1 chance in 1e8 seemed improbably small to me, but I did a quick check in R and got an even smaller number (2e-17) assuming a mean of 211 and an observation of 344.
ppois(344, lambda=211, lower=FALSE)
Using the 5 year mean and 2015 numbers from ic1000's link I saw:
ppois(342, lambda=222.6, lower=FALSE) = 5e-14
I think it would be better to use International Jew's method with a 5 year prior mean and current observation (rather than computing the mean of the prior and current year), but I think just using the prior year and current year would be good enough.
That would give (55.4 – 33.8) / sqrt((33) = 3.8 SD change. So a bit larger than the earlier 3.3 SD estimate, but not that much.Agreed. I picked Baltimore because it seems so blatantly nonrandom, yet the Wheeler paper seemed to question even it. Can you comment on the analysis in that paper?
Could you elaborate on how you would change International Jew's comment 18 to reflect both individual city and overall probability the change was nonrandom? I think his method his good for assessing the magnitude of the change (do you agree?), but the map to chance the change was random is imperfect because of your sample size point.
Thanks for your response!
P.S. Regarding rates vs. absolute counts, I'm not sure whether the more important effect to address is changing population size or small sample noise. I think the best approach would be to address both, but that makes things more difficult. How would you recommend doing that?Replies: @James B. Shearer, @James B. Shearer, @James B. Shearer
“Could you elaborate on this method? Is it commonly used? The 1 chance in 1e8 seemed improbably small to me, but I did a quick check in R and got an even smaller number (2e-17) assuming a mean of 211 and an observation of 344. …”
You should be using 555/2 =277.5 as the mean.
ppois(344, lambda=277.5, lower=FALSE) = 5.1e-5
which is a fair bit different from 1e-8, but much closer.
That said, do you have any further comments? It seems more intuitively reasonable to me to use the preceding 5 year mean (or perhaps include the current year as well in that) to test whether the current year is a nonrandom departure from usual.Replies: @James B. Shearer
Per a Wikipedia summary that pops up when you ask the question about % of black population in Dallas & Houston:
Population in Houston is 2.267 million versus 1.300 million in Dallas per Wikipedia.
This can be confusing since Dallas is often lumped with Ft. Worth and the entire Metroplex which in total is larger than Houston. Dallas “metro” population is 2.636 million. Houston metro area (including Galveston & close suburbs) is 7.07 million (both 2019).
The non Dallas/Houston suburbs and nearby cities tend to be less black than the actual major cities themselves, though Galveston may be an exception.
Still, it's remarkable that a city which is among the most black in the state of Texas has not seen a marked increase in its homicide rate over the last year while so many other less black cities in Texas have large increases in murders.Replies: @Muggles
https://statisticalatlas.com/place/Texas/Dallas/Race-and-Ethnicity
https://statisticalatlas.com/place/Texas/Houston/Race-and-Ethnicity
They give Dallas as 24.6% black with Houston as 22.8%.
But this page says 17.0% for Houston in 2018.
https://www.houston.org/houston-data/demographics-raceethnicity
This document says that in 2010 blacks were 16.8% in Houston and 15 percent in Dallas (and also compares with much smaller black populations in San Antonio, Austin, and El Paso).
This can be confusing since Dallas is often lumped with Ft. Worth and the entire Metroplex which in total is larger than Houston. Dallas "metro" population is 2.636 million. Houston metro area (including Galveston & close suburbs) is 7.07 million (both 2019).
The non Dallas/Houston suburbs and nearby cities tend to be less black than the actual major cities themselves, though Galveston may be an exception.Replies: @Pincher Martin, @res
Interesting. I thought Dallas was blacker than Houston – as a percentage of total population, of course, not in total numbers. But they’re about the same. Thanks for the info.
Still, it’s remarkable that a city which is among the most black in the state of Texas has not seen a marked increase in its homicide rate over the last year while so many other less black cities in Texas have large increases in murders.
Most murders are either family disputes/'romantic' problems or drug related. Those are pretty constant. Many people are armed and you can assume they are even if not.
I think most of the criminal drug crime is Hispanic based over territory or debts. Black population skews older and thus less homicidal than Hispanics.
The big bump in black murders was when the Katrina New Orleans black refugees headed to Houston. Didn't last long since Texas has much tougher sentencing for violent crimes. Also, braggart NO gangsters got wiped out quickly by the locals, who didn't like new competition.Replies: @Pincher Martin
You should be using 555/2 =277.5 as the mean.Replies: @res
OK, that gives:
ppois(344, lambda=277.5, lower=FALSE) = 5.1e-5
which is a fair bit different from 1e-8, but much closer.
That said, do you have any further comments? It seems more intuitively reasonable to me to use the preceding 5 year mean (or perhaps include the current year as well in that) to test whether the current year is a nonrandom departure from usual.
ppois(344, lambda=277.5, lower=FALSE) = 5.1e-5
which is a fair bit different from 1e-8, but much closer."
First to be super picky you need to use 343 not 344 so as to include 344.
Second you need to multiply by ppois(211,lambda=277.5) as you now have 2 unlikely events, a low year and a high year. When you do this you end up low at a bit over 1e-9. Part of the explanation is you are missing cases like 210 homicides in 2014 and 343 homicides in 2015 which would also be considered surprising. Also I think the binomial test is just better.Replies: @res
This can be confusing since Dallas is often lumped with Ft. Worth and the entire Metroplex which in total is larger than Houston. Dallas "metro" population is 2.636 million. Houston metro area (including Galveston & close suburbs) is 7.07 million (both 2019).
The non Dallas/Houston suburbs and nearby cities tend to be less black than the actual major cities themselves, though Galveston may be an exception.Replies: @Pincher Martin, @res
These pages have detailed statistical information (including interesting age/race breakdowns) for each city.
https://statisticalatlas.com/place/Texas/Dallas/Race-and-Ethnicity
https://statisticalatlas.com/place/Texas/Houston/Race-and-Ethnicity
They give Dallas as 24.6% black with Houston as 22.8%.
But this page says 17.0% for Houston in 2018.
https://www.houston.org/houston-data/demographics-raceethnicity
This document says that in 2010 blacks were 16.8% in Houston and 15 percent in Dallas (and also compares with much smaller black populations in San Antonio, Austin, and El Paso).
On what planet do blacks and hispanics copy whites? In real life, it’s exactly the opposite. Random counter-examples, off the top of my head: Colonel Parker bragged that in Elvis Presley he had found the ‘white negro’ he could use to sell black music. The Rolling Stones hit it big as some English guys copying R&B. The mass communication industries are busy replacing standard English with hoodrat patois. Pro sports are still dominated by black thugs. Etc., etc. ‘Blacks and hispanics copying whites, as usual.’ Go home and sleep it off.
The map of murder by county is rather silly, especially in the case of California. Del Norte County has a higher rate than neighboring county simply because it has a SuperMax prison. This is also true of some counties in other parts of the state. It’s murders by city people locked up in the country.
If you have red blood in your veins, this'll make it boil.
https://www.takimag.com/article/innocent-until-proven-trump-supporter/Replies: @Charon, @El Dato, @SunBakedSuburb, @Glt, @Ben tillman
It’s worse than that. His 90-IQ cousin who has a degree in Whites Are Evil spoke gibberish to the grand jury in an effort to convince them to indict. Part of her spiel was that her family is evil.
Still, it's remarkable that a city which is among the most black in the state of Texas has not seen a marked increase in its homicide rate over the last year while so many other less black cities in Texas have large increases in murders.Replies: @Muggles
Houston has a black mayor and Dallas had a black police chief for a long time. There were no major BLM riots in either city, though some small demonstrations. Houston’s mayor Turner is pretty law and order, as is the local population.
Most murders are either family disputes/’romantic’ problems or drug related. Those are pretty constant. Many people are armed and you can assume they are even if not.
I think most of the criminal drug crime is Hispanic based over territory or debts. Black population skews older and thus less homicidal than Hispanics.
The big bump in black murders was when the Katrina New Orleans black refugees headed to Houston. Didn’t last long since Texas has much tougher sentencing for violent crimes. Also, braggart NO gangsters got wiped out quickly by the locals, who didn’t like new competition.
Lubbock - 144% (8.6%)Fort Worth - 85% (18.9%)Arlington - 46% (18.8%)Austin - 43% (8.1%)Houston - 30% (25.7%)National average of all 59 US cities with data available - 28%San Antonio - 26% (6.9%)Dallas - 4% (24.7%)Plano - 0% (7.6%)So Houston doesn't stand out as being all that remarkable. Dallas does.
Most murders are either family disputes/'romantic' problems or drug related. Those are pretty constant. Many people are armed and you can assume they are even if not.
I think most of the criminal drug crime is Hispanic based over territory or debts. Black population skews older and thus less homicidal than Hispanics.
The big bump in black murders was when the Katrina New Orleans black refugees headed to Houston. Didn't last long since Texas has much tougher sentencing for violent crimes. Also, braggart NO gangsters got wiped out quickly by the locals, who didn't like new competition.Replies: @Pincher Martin
But Houston has seen an increase in homicides over the last year which is in line with the national average of all the cities in the study. Dallas has not.
I’ve put the black population for the Texas cities in the study both in parentheses and bold next to the increase in their homicide rates.
Lubbock – 144% (8.6%)
Fort Worth – 85% (18.9%)
Arlington – 46% (18.8%)
Austin – 43% (8.1%)
Houston – 30% (25.7%)
National average of all 59 US cities with data available – 28%
San Antonio – 26% (6.9%)
Dallas – 4% (24.7%)
Plano – 0% (7.6%)
So Houston doesn’t stand out as being all that remarkable. Dallas does.
I live a half hour from Plano, IL.
Stick, nice analogy. Maybe running out of ammo. A lot of these inner city shooting involve lots of shots, few mortal wounds.
Jon, 30,000 plus car break ins in San Francisco in one year and single digit arrests. People get tired of taking the time off when they know the out come. Shoplifting under $1000 isn’t even a crime now in some cities. You get an appearance ticket. The 8% drop probably means fewer crimes are reported.
https://stats.stackexchange.com/questions/15371/how-to-calculate-a-confidence-level-for-a-poisson-distributionCould you elaborate on this method? Is it commonly used? The 1 chance in 1e8 seemed improbably small to me, but I did a quick check in R and got an even smaller number (2e-17) assuming a mean of 211 and an observation of 344.
ppois(344, lambda=211, lower=FALSE)
Using the 5 year mean and 2015 numbers from ic1000's link I saw:
ppois(342, lambda=222.6, lower=FALSE) = 5e-14
I think it would be better to use International Jew's method with a 5 year prior mean and current observation (rather than computing the mean of the prior and current year), but I think just using the prior year and current year would be good enough.
That would give (55.4 – 33.8) / sqrt((33) = 3.8 SD change. So a bit larger than the earlier 3.3 SD estimate, but not that much.Agreed. I picked Baltimore because it seems so blatantly nonrandom, yet the Wheeler paper seemed to question even it. Can you comment on the analysis in that paper?
Could you elaborate on how you would change International Jew's comment 18 to reflect both individual city and overall probability the change was nonrandom? I think his method his good for assessing the magnitude of the change (do you agree?), but the map to chance the change was random is imperfect because of your sample size point.
Thanks for your response!
P.S. Regarding rates vs. absolute counts, I'm not sure whether the more important effect to address is changing population size or small sample noise. I think the best approach would be to address both, but that makes things more difficult. How would you recommend doing that?Replies: @James B. Shearer, @James B. Shearer, @James B. Shearer
“Agreed. I picked Baltimore because it seems so blatantly nonrandom, yet the Wheeler paper seemed to question even it. Can you comment on the analysis in that paper?”
It appears to me that the paper is not claiming that the change is nonrandom just that it is not dramatically larger than other nonrandom changes in Baltimore in the past.
Actually, parsing their words more carefully, I agree with this part of what they said in the abstract:Both of those are rather weak statements and have been spun as being far stronger in the popular press.
I also question whether "other nonrandom changes in Baltimore in the past" is the appropriate comparison metric. Is anyone seriously going to argue that "not as bad as what happened during the crack epidemic" is acceptable?
ppois(344, lambda=277.5, lower=FALSE) = 5.1e-5
which is a fair bit different from 1e-8, but much closer.
That said, do you have any further comments? It seems more intuitively reasonable to me to use the preceding 5 year mean (or perhaps include the current year as well in that) to test whether the current year is a nonrandom departure from usual.Replies: @James B. Shearer
“OK, that gives:
ppois(344, lambda=277.5, lower=FALSE) = 5.1e-5
which is a fair bit different from 1e-8, but much closer.”
First to be super picky you need to use 343 not 344 so as to include 344.
Second you need to multiply by ppois(211,lambda=277.5) as you now have 2 unlikely events, a low year and a high year. When you do this you end up low at a bit over 1e-9. Part of the explanation is you are missing cases like 210 homicides in 2014 and 343 homicides in 2015 which would also be considered surprising. Also I think the binomial test is just better.
https://stats.stackexchange.com/questions/15371/how-to-calculate-a-confidence-level-for-a-poisson-distributionCould you elaborate on this method? Is it commonly used? The 1 chance in 1e8 seemed improbably small to me, but I did a quick check in R and got an even smaller number (2e-17) assuming a mean of 211 and an observation of 344.
ppois(344, lambda=211, lower=FALSE)
Using the 5 year mean and 2015 numbers from ic1000's link I saw:
ppois(342, lambda=222.6, lower=FALSE) = 5e-14
I think it would be better to use International Jew's method with a 5 year prior mean and current observation (rather than computing the mean of the prior and current year), but I think just using the prior year and current year would be good enough.
That would give (55.4 – 33.8) / sqrt((33) = 3.8 SD change. So a bit larger than the earlier 3.3 SD estimate, but not that much.Agreed. I picked Baltimore because it seems so blatantly nonrandom, yet the Wheeler paper seemed to question even it. Can you comment on the analysis in that paper?
Could you elaborate on how you would change International Jew's comment 18 to reflect both individual city and overall probability the change was nonrandom? I think his method his good for assessing the magnitude of the change (do you agree?), but the map to chance the change was random is imperfect because of your sample size point.
Thanks for your response!
P.S. Regarding rates vs. absolute counts, I'm not sure whether the more important effect to address is changing population size or small sample noise. I think the best approach would be to address both, but that makes things more difficult. How would you recommend doing that?Replies: @James B. Shearer, @James B. Shearer, @James B. Shearer
“P.S. Regarding rates vs. absolute counts, I’m not sure whether the more important effect to address is changing population size or small sample noise. I think the best approach would be to address both, but that makes things more difficult. How would you recommend doing that?”
It is easy to adjust my proposed test for a changing population size (as long as you know the sizes), just use a binomial distribution with p other than .5.
binom.test(211, 555) giving p-value = 1.811e-08
Right?
If we assume the population increased 10% between years is the new test just this?
binom.test(211, 555, p = 1 / 2.1) giving p-value = 6.251e-06
Perhaps I am being dense, but it is not immediately obvious to me why that would work. My intuition is to do something more complex like
binom.test(211, round(211 + 344 / 1.1)) giving p-value = 9.629e-06
It just seems like the 555 should be corrected for the different population sizes as well. I think you can see that in the large change limit.
But then again, that is not needed as long as the population changes are small. Is that the right way to think about it?
Is there any way to expand that test to include multiple prior years?
Thanks for your patience in explaining all of this!Replies: @James B. Shearer
I wonder why murders are down slightly, 2020 cf. 2019.Replies: @stillCARealist, @Mark Roulo, @Tony, @Ga On My Mind
The murder rate of Baltimore is down slightly for one reason only: the previous year (2019) was the highest murder rate in the city’s entire history. Search “2019 Baltimore murder rate”:
“Baltimore ended 2019 with 348 homicides on record, according to data compiled by The Baltimore Sun. The year had already set a grim record of 57 killings per 100,000 people, the city’s worst homicide rate on record.”
It had almost nowhere to go but down.
Gopnick succeeded in writing a very, very long article about nothing.
Even in its ostensibly non-political coverage, The NY Times really has become pathetic. Even the article’s commenters make fun of this lame attempt to say something that isn’t a) obvious, b) trivial, or c) transparently partisan.
I understand that a young Francis Bacon taught himself to paint by copying Picasso and studying the photographs of Eadweard Muybridge. Another reputation down the drain.
It appears to me that the paper is not claiming that the change is nonrandom just that it is not dramatically larger than other nonrandom changes in Baltimore in the past.Replies: @res
Agreed. “Question even it” was poor wording. Minimize would have been better given how obviously large the change was by the metrics we are discussing. My overall concern is more about the judgment that the overall (nationwide) effect was not nonrandom. I should have been clearer that I was asking for comment on the full paper analysis, not just Baltimore.
Actually, parsing their words more carefully, I agree with this part of what they said in the abstract:
Both of those are rather weak statements and have been spun as being far stronger in the popular press.
I also question whether “other nonrandom changes in Baltimore in the past” is the appropriate comparison metric. Is anyone seriously going to argue that “not as bad as what happened during the crack epidemic” is acceptable?
ppois(344, lambda=277.5, lower=FALSE) = 5.1e-5
which is a fair bit different from 1e-8, but much closer."
First to be super picky you need to use 343 not 344 so as to include 344.
Second you need to multiply by ppois(211,lambda=277.5) as you now have 2 unlikely events, a low year and a high year. When you do this you end up low at a bit over 1e-9. Part of the explanation is you are missing cases like 210 homicides in 2014 and 343 homicides in 2015 which would also be considered surprising. Also I think the binomial test is just better.Replies: @res
Thanks for the clarification. I forgot lower.tail is inclusive and upper.tail is exclusive.
Thanks. That is why I prefer looking at a five year average for the mean instead.
Fair enough. Can you offer any thoughts on what the range of validity is? For example, how few events in a year allow it?
You can do the test with any number of events but the smaller the number of events the bigger the change has to be to stand out.
In the case at hand assume rates of 34 and 55 per 100000. Then if the population is 100000 you have 34 and 55 homicides and the (one-sided) test is pbinom(34,89,.5) = .0167 which isn't that impressive. Increase the population to 200,000 and we have pbinom(68,178,.5) = .001. Increase the population to 400000 and we have pbinom(136,356,.5) = 5e-6. To 600,000 (close to Baltimore's actual population) and we have pbinom(204,534,.5) = 2.8e-8. To 1,000,000 and we have pbinom(340,890,.5) = 9.76e-13.
So it looks you need a city population of 200,000 or so before a change like this between 2 years starts to really stand out.Replies: @res
It is easy to adjust my proposed test for a changing population size (as long as you know the sizes), just use a binomial distribution with p other than .5.Replies: @res
Thanks. To make sure I understand, here is your original statement.
This is equivalent to
binom.test(211, 555) giving p-value = 1.811e-08
Right?
If we assume the population increased 10% between years is the new test just this?
binom.test(211, 555, p = 1 / 2.1) giving p-value = 6.251e-06
Perhaps I am being dense, but it is not immediately obvious to me why that would work. My intuition is to do something more complex like
binom.test(211, round(211 + 344 / 1.1)) giving p-value = 9.629e-06
It just seems like the 555 should be corrected for the different population sizes as well. I think you can see that in the large change limit.
But then again, that is not needed as long as the population changes are small. Is that the right way to think about it?
Is there any way to expand that test to include multiple prior years?
Thanks for your patience in explaining all of this!
binom.test(211, 555) giving p-value = 1.811e-08
Right?"I would use pbinom(211,555,.5) = 9.05e-09. Multiply by 2 to get your number which is apparently a two-sided test. "If we assume the population increased 10% between years is the new test just this?
binom.test(211, 555, p = 1 / 2.1) giving p-value = 6.251e-06"I would use pbinom(211,555,1/2.1) = 3.2e-6. I am not sure what binom.test is doing when p isn't .5. If the population decreased 10% I would use pbinom(211,555,1/1.9) = 3.3e-12."It just seems like the 555 should be corrected for the different population sizes as well. I think you can see that in the large change limit."Think of it like lottery tickets. Suppose each person each year buys a ticket and then after two years we draw 555 tickets. How likely are you to draw 211 tickets or fewer from the first year. Note we are assuming the population is big enough that we don't have to worry about the difference between drawing with or without replacement."Is there any way to expand that test to include multiple prior years?"Sure, just think of selling lottery tickets to one person each year and then drawing a number of tickets equal to the total number of homicides and ask how likely the last year was likely to draw that many tickets (homicides) or more (if you are testing for an increase).This is all assuming of course that homicides occur one at a time and independently. The existence of multiple homicide events and revenge killings will make the data noisier and make changes in the rate less significant.
“Fair enough. Can you offer any thoughts on what the range of validity is? For example, how few events in a year allow it?”
You can do the test with any number of events but the smaller the number of events the bigger the change has to be to stand out.
In the case at hand assume rates of 34 and 55 per 100000. Then if the population is 100000 you have 34 and 55 homicides and the (one-sided) test is pbinom(34,89,.5) = .0167 which isn’t that impressive. Increase the population to 200,000 and we have pbinom(68,178,.5) = .001. Increase the population to 400000 and we have pbinom(136,356,.5) = 5e-6. To 600,000 (close to Baltimore’s actual population) and we have pbinom(204,534,.5) = 2.8e-8. To 1,000,000 and we have pbinom(340,890,.5) = 9.76e-13.
So it looks you need a city population of 200,000 or so before a change like this between 2 years starts to really stand out.
https://stat.ethz.ch/pipermail/r-help/2012-August/333722.html
That mentions two issues. First, that the two sided test result is not always 2x the one sided test. Second, that the upper tail (not relevant to your comment) is inclusive for binom.test and exclusive for pbinom. Which seems odd given that the option for binom.test is called "greater".So the sample size effect overwhelms any issue arising from differences between the Poisson and binomial distributions?
binom.test(211, 555) giving p-value = 1.811e-08
Right?
If we assume the population increased 10% between years is the new test just this?
binom.test(211, 555, p = 1 / 2.1) giving p-value = 6.251e-06
Perhaps I am being dense, but it is not immediately obvious to me why that would work. My intuition is to do something more complex like
binom.test(211, round(211 + 344 / 1.1)) giving p-value = 9.629e-06
It just seems like the 555 should be corrected for the different population sizes as well. I think you can see that in the large change limit.
But then again, that is not needed as long as the population changes are small. Is that the right way to think about it?
Is there any way to expand that test to include multiple prior years?
Thanks for your patience in explaining all of this!Replies: @James B. Shearer
“This is equivalent to
binom.test(211, 555) giving p-value = 1.811e-08
Right?”
I would use pbinom(211,555,.5) = 9.05e-09. Multiply by 2 to get your number which is apparently a two-sided test.
“If we assume the population increased 10% between years is the new test just this?
binom.test(211, 555, p = 1 / 2.1) giving p-value = 6.251e-06”
I would use pbinom(211,555,1/2.1) = 3.2e-6. I am not sure what binom.test is doing when p isn’t .5. If the population decreased 10% I would use pbinom(211,555,1/1.9) = 3.3e-12.
“It just seems like the 555 should be corrected for the different population sizes as well. I think you can see that in the large change limit.”
Think of it like lottery tickets. Suppose each person each year buys a ticket and then after two years we draw 555 tickets. How likely are you to draw 211 tickets or fewer from the first year. Note we are assuming the population is big enough that we don’t have to worry about the difference between drawing with or without replacement.
“Is there any way to expand that test to include multiple prior years?”
Sure, just think of selling lottery tickets to one person each year and then drawing a number of tickets equal to the total number of homicides and ask how likely the last year was likely to draw that many tickets (homicides) or more (if you are testing for an increase).
This is all assuming of course that homicides occur one at a time and independently. The existence of multiple homicide events and revenge killings will make the data noisier and make changes in the rate less significant.
You can do the test with any number of events but the smaller the number of events the bigger the change has to be to stand out.
In the case at hand assume rates of 34 and 55 per 100000. Then if the population is 100000 you have 34 and 55 homicides and the (one-sided) test is pbinom(34,89,.5) = .0167 which isn't that impressive. Increase the population to 200,000 and we have pbinom(68,178,.5) = .001. Increase the population to 400000 and we have pbinom(136,356,.5) = 5e-6. To 600,000 (close to Baltimore's actual population) and we have pbinom(204,534,.5) = 2.8e-8. To 1,000,000 and we have pbinom(340,890,.5) = 9.76e-13.
So it looks you need a city population of 200,000 or so before a change like this between 2 years starts to really stand out.Replies: @res
Thanks. Here is some discussion of pbinom vs. binom.test
https://stat.ethz.ch/pipermail/r-help/2012-August/333722.html
That mentions two issues. First, that the two sided test result is not always 2x the one sided test. Second, that the upper tail (not relevant to your comment) is inclusive for binom.test and exclusive for pbinom. Which seems odd given that the option for binom.test is called “greater”.
So the sample size effect overwhelms any issue arising from differences between the Poisson and binomial distributions?