Chance News 111: Difference between revisions
No edit summary |
|||
(120 intermediate revisions by the same user not shown) | |||
Line 1: | Line 1: | ||
July 1, 2017 to August 20, 2017 | |||
==Quotations== | ==Quotations== | ||
“In my darkest moods I follow what could be called the ‘Groucho principle’: | “In my darkest moods I follow what could be called the ‘Groucho principle’: | ||
Line 12: | Line 12: | ||
<div align=right>-- Paul Alper, in: Occom, Mencken and Cohn, ''Higher Education Review'', vol. 49, no.2 </div> | <div align=right>-- Paul Alper, in: Occom, Mencken and Cohn, ''Higher Education Review'', vol. 49, no.2 </div> | ||
---- | |||
"Indeed, the statistician David Freedman used to say that if the topic of regression comes up in a criminal or civil trial, the side that must explain regression to the jury will lose the case." | |||
<div align=right> -- Daniel Kahneman, in: ''Thinking, Fast and Slow'' (Farrar, Straus, and Giroux, 2013), p. 182.</div> | |||
==Forsooth== | ==Forsooth== | ||
Line 30: | Line 33: | ||
Photo by Pete Schumer, taken in the elevator at his hotel. | Photo by Pete Schumer, taken in the elevator at his hotel. | ||
---- | |||
Too much precision! | |||
"By 2025, the [National Funeral Directors] Association is forecasting that 63.8 percent of the people who die in the United States will be cremated, and by 2035, 78.8 percent." | |||
<div align=right>in: [https://www.nytimes.com/2017/08/10/nyregion/cremations-increase-in-a-move-away-from-tradition.html In a move away from tradition, cremations increase], ''New York Times'', 10 August 2017</div> | |||
Submitted by Paul Alper | |||
=="Random" seat assignment== | =="Random" seat assignment== | ||
Line 51: | Line 62: | ||
cast doubt on whether Ryanair's seat allocation can be purely random." | cast doubt on whether Ryanair's seat allocation can be purely random." | ||
The article concludes with the following explanation from Ryanair, which | The article concludes with the following explanation from Ryanair, which may qualify as an extended Forsooth! | ||
<blockquote> | <blockquote> | ||
We haven't changed the random seat | We haven't changed the random seat | ||
allocation policy. | allocation policy.''' | ||
<br><br> | <br><br> | ||
The reason for more middle seats being allocated is that more and more | The reason for more middle seats being allocated is that more and more | ||
Line 72: | Line 83: | ||
aisle seats free to facilitate those customers who are willing to pay | aisle seats free to facilitate those customers who are willing to pay | ||
(from £2) for them. | (from £2) for them. | ||
</blockquote> | </blockquote>''' | ||
Submitted by Patrick O'Beirne | Submitted by Patrick O'Beirne | ||
===Discussion=== | |||
#Emil Friedman wrote to say that the airline's explanation makes sense if we understand timing: the random assignment will be made after the reserved seats are fixed. Should the advertisement be reworded to reflect this? | |||
#Should there be some statement of (historical) odds, as on the lottery web page? How do you think the Oxford team estimated the odds? | |||
==Sense and Sensibility and Statistics== | ==Sense and Sensibility and Statistics== | ||
Line 90: | Line 105: | ||
Submitted by Steve Simon | Submitted by Steve Simon | ||
===Followup=== | |||
[https://www.nytimes.com/2017/07/15/insider/charting-literary-greatness-with-jane-austen.html Charting Literary Greatness With Jane Austen]<br> | |||
by Kathleen A. Flynn, ''New York Times'' "Times Insider", 15 July 2017 | |||
Flynn, one of the authors of the above article, describes on how the Austen project arose. Of the quantitative approach, she writes <blockquote> | |||
To reduce words to data points is not to diminish the power of choosing the right ones and putting them in the right order; there’s magic in fiction’s ability to explore universal human truths. But a discipline like statistics can offer a new and amusing — even instructive — look at something as intuitive and artistic as a novel, and that’s exciting, too. | |||
</blockquote> | |||
The aritcle provides links to other data-analytic studies of literature. For example, scholars at [http://austen.unl.edu/about#team University of Nebraska-Lincoln] have computed word frequencies for all of Austen's works, while the [http://litlab.stanford.edu/pamphlets/ Stanford Literacy Project] has applied "computational criticism" to a wide range of literary analyses. | |||
==Do the ''stats''!== | |||
[http://www.chronicle.com/blogs/letters/statistical-errors-are-often-not-due-to-mathematical-errors/ Statistical errors are often not due to mathematical errors]<br> | |||
by Brian Zaharatos, letter to the editor, ''Chronicle of Higher Education'', 11 July 2017 | |||
Zaharatos writes in response to a June 28 article from the Chronicle, entitled [http://www.chronicle.com/article/A-New-Theory-on-How/240470 A new theory on how researchers can solve the reproducibility crisis: Do the math]. While acknowledging some good points raised in the article, he notes that it also misinterprets several statistical concepts, as in this faulty description of significance: | |||
<blockquote> | |||
[F]or a p-value of 0.05…a study’s finding will be deemed significant if researchers identify a 95-percent chance that it is genuine. | |||
</blockquote> | |||
Regarding the headline, notes that people's difficulties interpreting statistical studies are not purely mathematical: "Mathematical skills are necessary for success in statistics, but they are far from sufficient. Statisticians — and researchers using statistics — ought to have a nuanced understanding of the concepts that mathematics helps them quantify." | |||
==Gerrymandering== | ==Gerrymandering== | ||
Line 95: | Line 130: | ||
NBCnews.com, 25 June 2017 | NBCnews.com, 25 June 2017 | ||
Republican electoral gains in 2010 gave the party substantial control over the [http://redistricting.lls.edu/what.php redistricting] that followed the 2010 Census. NBC reports on a statistical analysis by the Associated Press showing the effect in subsequent elections and the state and national levels. For example, in the 2016 election, they estimate that Republicans won | Republican electoral gains in 2010 gave the party substantial control over the [http://redistricting.lls.edu/what.php redistricting] that followed the 2010 Census. NBC reports on a statistical analysis by the Associated Press showing the effect in subsequent elections and the state and national levels. For example, in the 2016 election, they estimate that "Republicans won as many as 22 additional U.S. House seats over what would have been expected based on the average vote share in congressional districts across the country." The AP website reports that [https://apnews.com/bfbf76adbcf8461eab9096ca7ab4cf83 Texas had the largest gain], nearly 4 extra seats. | ||
Gerrymandering is of course nothing new. The reference to Texas brought to mind a famous story, retold in a 2002 story from the ''Economist'' ([http://www.economist.com/node/1099030 How to rig an election], 25 April 2002): "[It] used to be said of the old Texas 6th (which was a road from Houston to Dallas), that you could kill most of the constituents by driving down the road with the car doors open." | |||
However, the latest efforts are seen as particularly egregious. We read, | |||
<blockquote> | <blockquote> | ||
The AP’s analysis was based on an “efficiency gap” formula developed by University of Chicago law professor Nick Stephanopoulos and Eric McGhee, a researcher at the nonpartisan Public Policy Institute of California. Their mathematical model was cited last fall as “corroborative evidence” by a federal appeals court panel that struck down Wisconsin’s Assembly districts as an intentional partisan gerrymander in violation of Democratic voters’ rights to representation. The U.S. Supreme Court has agreed to hear an appeal. | The AP’s analysis was based on an “efficiency gap” formula developed by University of Chicago law professor Nick Stephanopoulos and Eric McGhee, a researcher at the nonpartisan Public Policy Institute of California. Their mathematical model was cited last fall as “corroborative evidence” by a federal appeals court panel that struck down Wisconsin’s Assembly districts as an intentional partisan gerrymander in violation of Democratic voters’ rights to representation. The U.S. Supreme Court has agreed to hear an appeal. | ||
Line 108: | Line 145: | ||
</blockquote> | </blockquote> | ||
NBC | NBC News also cites another analysis approach, introduced by the [http://election.princeton.edu/2017/01/17/a-new-project-on-partisan-gerrymandering/ Princeton University Gerrymandering Project], launched this year by Sam Wang. The inaugural post includes this graphic | ||
<center>[[File:Princeton gerrymander.png|450px|frameless]]</center> | |||
which shows how the last three election cycles have been unusual. | |||
The | |||
Wang has proposed [https://web.math.princeton.edu/~sswang/wang16_ElectionLawJournal_gerrymandering-MD-WI_.pdf three statistical tests] for detecting gerrymandering. The Princeton Project site has a good description of [http://gerrymander.princeton.edu/info how the tests work], and provides for [http://gerrymander.princeton.edu/test interactive demonstrations]. | |||
The NBC News story summarized results from the above tests by saying, | |||
<blockquote> | |||
...the extreme Republican advantages in some states were no fluke. The Republican edge in Michigan's state House districts had only a 1-in-16,000 probability of occurring by chance; in Wisconsin's Assembly districts, there was a mere 1-in-60,000 likelihood of it happening randomly... | |||
</blockquote> | |||
These are p-values, which are often misinterpreted in news stories. (Chance News reader) Jeff Eiseman provided the following clarification. | |||
<blockquote> | |||
If the relationship between the Democrat’s share of national votes and the Democrat’s share of sheets in Michigan were consistent with those before 2012, the chances that the Republicans would have achieved an edge as large or larger than they did in Michigan and Wisconsin would be only 1 out of 16,000 and 1 out of 60,000 respectively. | |||
</blockquote> | |||
Submitted by Bill Peterson | |||
==More on mortality== | ==More on middle-age mortality== | ||
[http://andrewgelman.com/2017/07/11/criticism-economists-journalists-jumping-conclusions-based-mortality-trends/ Further criticism of social scientists and journalists jumping to conclusions based on mortality trends]<br> | [http://andrewgelman.com/2017/07/11/criticism-economists-journalists-jumping-conclusions-based-mortality-trends/ Further criticism of social scientists and journalists jumping to conclusions based on mortality trends]<br> | ||
by Andrew Gelman, "Statistical Modeling, Causal Inference, and Social Science" blog, 11 July 2017 | by Andrew Gelman, "Statistical Modeling, Causal Inference, and Social Science" blog, 11 July 2017 | ||
Line 121: | Line 170: | ||
In [https://www.causeweb.org/wiki/chance/index.php/Chance_News_110#Deaths_of_despair Deaths of despair] from the last installment of Chance News, we described the Angus-Deaton work on middle-age mortality, including commentary by Gelman and others. In this latest post, Gelman includes links to some other articles suggested by his readers. | In [https://www.causeweb.org/wiki/chance/index.php/Chance_News_110#Deaths_of_despair Deaths of despair] from the last installment of Chance News, we described the Angus-Deaton work on middle-age mortality, including commentary by Gelman and others. In this latest post, Gelman includes links to some other articles suggested by his readers. | ||
Gelman reiterates his support for the project, which he | Gelman reiterates his support for the project, which he believes has raised important issues. But he points out that the omission age-adjustment was strange, since | ||
"reporting trends in mortality rates without age adjusting is like reporting trends in nominal prices without adjusting for the consumer price index." | "reporting trends in mortality rates without age adjusting is like reporting trends in nominal prices without adjusting for the consumer price index." | ||
Line 131: | Line 180: | ||
by Barry D. Nussbaum and Katherine K. Wallman, ''Huffington Post'', 7 July 2017 | by Barry D. Nussbaum and Katherine K. Wallman, ''Huffington Post'', 7 July 2017 | ||
The authors of this op-ed are present and past presidents of the American Statistical Association. They describe how | The authors of this op-ed are present and past presidents of the American Statistical Association. They describe how the former chief statistician of Greece, Andreas Georgiou, is being prosecuted for his reporting on the Greek national debt. The issue is not his work, which has actually produced a more accurate accounting, but rather the revelation that the debt is larger than previously understood, which will require stricter austerity measures. Nussbaum and Wallman call for US government officials to stress the need for transparent statistical data, and more generally to condemn attacks on science. | ||
In the US, statisticians are not being prosecuted for their analyses. However, the Greek saga calls to mind how climate scientists have been [https://www.theguardian.com/environment/2017/feb/22/climate-change-science-attacks-threats-trump subjected to harrassment]. Indeed, in February Representative Lamar Smith (R-Texas), chair of House Science Committee, convened hearings dubiously titled [https://arstechnica.com/science/2017/02/house-science-committee-holds-hearing-on-making-epa-great-again/ “Making EPA Great Again”]. | In the US, statisticians are not being prosecuted for their analyses. However, the Greek saga calls to mind how climate scientists have been [https://www.theguardian.com/environment/2017/feb/22/climate-change-science-attacks-threats-trump subjected to harrassment]. Indeed, in February Representative Lamar Smith (R-Texas), chair of House Science Committee, convened hearings dubiously titled [https://arstechnica.com/science/2017/02/house-science-committee-holds-hearing-on-making-epa-great-again/ “Making EPA Great Again”]. | ||
In another story, the director of the US Census Bureau John Thompson announced his resignation this May. | In another recent story, the director of the US Census Bureau John Thompson announced his resignation this May. During an interview with NPR, [http://www.npr.org/sections/codeswitch/2017/07/15/536908867/could-a-census-without-a-leader-spell-trouble-in-2020 Could a census without a leader spell trouble In 2020?] (NPR "Code Switch", 15 July 2017), Thompson expressed confidence that work at the bureau would continue its work without disruption. Some observers have expressed concern because the 2020 census is fast approaching. Allocation of funds for many federal programs is | ||
[https://www.brookings.edu/wp-content/uploads/2016/06/0309_census_report.pdf dependent on census data]. The counts are also have implications for Congressional redistricting. The Trump adminstration has so far been slow at [https://www.nytimes.com/interactive/2017/07/17/us/politics/trump-appointments.html filling top jobs]. | [https://www.brookings.edu/wp-content/uploads/2016/06/0309_census_report.pdf dependent on census data]. The counts are also have implications for Congressional redistricting. The Trump adminstration has so far been slow at [https://www.nytimes.com/interactive/2017/07/17/us/politics/trump-appointments.html filling top jobs]. | ||
Submitted by Bill Peterson | |||
===ASA statement on Georgiou case=== | |||
ASA August Member News [http://www.amstat.org/ASA/News/Georgiou-Convicted-Again-the-ASA-Continues-its-Pledge.aspx? reports] that Georgiou has been convicted by an appeals court in Greece. Earlier this month, Georgiou delivered a presentation on [http://ww2.amstat.org/meetings/jsm/2017/onlineprogram/mobile/typedetails.cfm?SessionID=214595&type=1&subtype=8 National Governments, Coerced Narratives, Creative Language, and Alternative Facts] at the Joint Statistical Meetings. The ASA has issued the following statement: | |||
<blockquote> | |||
We were disturbed to learn this morning of the conviction of Andreas Georgiou, the former president of the Hellenic Statistical Authority (ELSTAT), on charges of violation of duty by an appeals court in Greece. Dr. Georgiou was acquitted of these charges in December 2015; the decision to retry him amounts to double-jeopardy. As noted in our [http://www.amstat.org/asa/files/pdfs/POL-ASALetterRegardingAndreasGeorgiou2017-03-27.pdf March 27 letter] to Alexis Tsipras, Prime Minister of the Hellenic Republic, “There is widespread agreement among statistical authorities that Dr. Georgiou compiled Greek public finance statistics in accordance with EU rules and standards, with which Greece is obliged to comply, and that the charges against him are baseless.” We reiterate the call that: “all remaining charges against Dr. Georgiou and his former colleagues be dropped immediately. Under no circumstances can criminal sanctions for developing and presenting independent and objective statistical data in a credible way be justified." | |||
</blockquote> | |||
==Still running against Hillary?== | ==Still running against Hillary?== | ||
Line 144: | Line 201: | ||
:by Philip Bump, ''Washington Post'', 21 July 2017 | :by Philip Bump, ''Washington Post'', 21 July 2017 | ||
Political observers have noted that President Trump thrives on [https://www.nytimes.com/2017/02/16/us/politics/campaign-over-president-trump-will-hold-a-what-else-campaign-rally.html campaign-style events], and that his communications regularly lapse into campaign rhetoric. The ''Post'' presents a quantitative analysis of how often Trump invokes the name of his 2016 opponent, Hillary Clinton, and other terms relating to the election. The following graphic from the article shows | Political observers have noted that President Trump thrives on [https://www.nytimes.com/2017/02/16/us/politics/campaign-over-president-trump-will-hold-a-what-else-campaign-rally.html campaign-style events], and that his communications regularly lapse into campaign rhetoric. The ''Post'' presents a quantitative analysis of how often Trump invokes the name of his 2016 opponent, Hillary Clinton, and other terms relating to the election. The following graphic from the article shows how many of Trump's 19 interviews to date have included such references, and how far into the interview the first reference appears. | ||
[[File:How far.jpeg|400px|center]] | [[File:How far.jpeg|400px|center]] | ||
Line 154: | Line 211: | ||
[[File:How far_Hillary.jpeg|450px|center]] | [[File:How far_Hillary.jpeg|450px|center]] | ||
Links to the interviews are provided at the end of the article. Quoting the beginning of the [https://www.nytimes.com/2017/07/19/us/politics/trump-interview-transcript.html transcript] of that last interview from the ''Times'': | Similar graphs are constructed for the other terms. | ||
Links to the interviews themselves are provided at the end of the article. Quoting the beginning of the [https://www.nytimes.com/2017/07/19/us/politics/trump-interview-transcript.html transcript] of that last interview from the ''Times'': | |||
<blockquote> | <blockquote> | ||
TRUMP: Hi fellas, how you doing? | TRUMP: Hi fellas, how you doing? | ||
Line 162: | Line 221: | ||
TRUMP: It was good. We are very close. It’s a tough — you know, health care. Look, ''Hillary Clinton'' worked eight years in the White House with her husband as president and having majorities and couldn’t get it done. Smart people, tough people — couldn’t get it done. ''Obama'' worked so hard... (''emphasis added'') | TRUMP: It was good. We are very close. It’s a tough — you know, health care. Look, ''Hillary Clinton'' worked eight years in the White House with her husband as president and having majorities and couldn’t get it done. Smart people, tough people — couldn’t get it done. ''Obama'' worked so hard... (''emphasis added'') | ||
</blockquote> | </blockquote> | ||
As the ''Post'' reported, he got though just 20 words before "Hillary." The caption for the second graphic says that the figures are based on "character analysis." This is not explicitly defined in the article, but the preceding suggests that it means | As the ''Post'' reported, we can see that he got though just 20 words before "Hillary." The caption for the second graphic says that the figures are based on "character analysis." This is not explicitly defined in the article, but the preceding suggests that it means word counts, calculating the number of words he spoke before the first "Hillary" divided by his total words in the interview. | ||
'''Discussion'''<br> | '''Discussion'''<br> | ||
Line 170: | Line 229: | ||
==Voter fraud and the birthday problem== | ==Voter fraud and the birthday problem== | ||
[https://www.washingtonpost.com/news/wonk/wp/2017/07/20/this-anti-voter-fraud-program-gets-it-wrong-over-99-of-the-time-the-gop-wants-to-take-it-nationwide/?utm_term=.7028deb33941 This anti-voter-fraud program gets it wrong over 99 percent of the time. The GOP wants to take it nationwide.] <br> | [https://www.washingtonpost.com/news/wonk/wp/2017/07/20/this-anti-voter-fraud-program-gets-it-wrong-over-99-of-the-time-the-gop-wants-to-take-it-nationwide/?utm_term=.7028deb33941 This anti-voter-fraud program gets it wrong over 99 percent of the time. The GOP wants to take it nationwide.] <br> | ||
by Christopher Ingraham, ''Washington Post'' | by Christopher Ingraham, "Wonkblog", ''Washington Post'', 20 July 2017 | ||
Throughout the 2016 campaign, candidate Trump warned that the election [https://twitter.com/realdonaldtrump/status/787699930718695425?lang=en was being rigged]. After the election, he complained that Hillary Clinton would not have won the popular vote without [http://www.reuters.com/article/us-usa-trump-votes-idUSKBN13M0XZ millions of illegal ballots]. Such claims have cast a shadow over the recently convened [http://www.politico.com/story/2017/07/19/trump-pence-voter-fraud-commission-240714 Presidential Advisory Commission on Election Integrity], despite Vice President Pence's assurances that "this commission has no preconceived notions." | Throughout the 2016 campaign, candidate Trump warned that the election [https://twitter.com/realdonaldtrump/status/787699930718695425?lang=en was being rigged]. After the election, he complained that Hillary Clinton would not have won the popular vote without [http://www.reuters.com/article/us-usa-trump-votes-idUSKBN13M0XZ millions of illegal ballots]. Such claims have cast a shadow over the recently convened [http://www.politico.com/story/2017/07/19/trump-pence-voter-fraud-commission-240714 Presidential Advisory Commission on Election Integrity], despite Vice President Pence's assurances that "this commission has no preconceived notions." | ||
Kansas Secretary of State Kris Kobach is the vice-chairman of the commission. As described in the ''Post'', he has advocated national implementation of Kansas program called [https://en.wikipedia.org/wiki/Interstate_Voter_Registration_Crosscheck_Program Interstate Crosscheck], which | Kansas Secretary of State Kris Kobach is the vice-chairman of the commission. As described in the ''Post'', he has advocated national implementation of Kansas program called [https://en.wikipedia.org/wiki/Interstate_Voter_Registration_Crosscheck_Program Interstate Voter Registration Crosscheck], which seeks to identify people | ||
who may be voting in more than one location. | |||
Other states are invited | |||
to send their voter registration lists to Kansas, and Crosscheck will search for voters with matching first name, last name and date of birth. | |||
The idea is to remove duplicate registrations, though it is not exactly clear how this would be implemented. | |||
Critics of the program argue that it creates many "false positives"; that is, different people with matching names and birthdays. From the article we read: | Critics of the program argue that it creates many "false positives"; that is, different people with matching names and birthdays. From the article we read: | ||
Line 182: | Line 245: | ||
</blockquote> | </blockquote> | ||
Ostensibly, Crosscheck appears to assume that a match on name and birthdate indicates a duplicate registration by the same person. Calculating the chance of such a match among ''different'' people is an application of a famous problem from probability. The ''Post'' cites a 2007 paper, [https://papers.ssrn.com/sol3/papers.cfm?abstract_id=997888 Seeing double voting: An extension of the birthday problem]. It is well-known that among 23 randomly selected birthdays (month and day), there is a better than even chance that at least two will match. | |||
With 180 people, the article reports that there is a 50% chance that two will match birthdates (month, day and year). There is a built-in function in R called [https://stat.ethz.ch/R-manual/R-devel/library/stats/html/birthday.html qbirthday] that implements such calculations. For example: | With 180 people, the article reports that there is a 50% chance that two will match birthdates (month, day and year). There is a built-in function in R called [https://stat.ethz.ch/R-manual/R-devel/library/stats/html/birthday.html qbirthday] that implements such calculations. For example: | ||
Line 192: | Line 255: | ||
</code> | </code> | ||
For the second calculation we assume that an individual has 65 years of voting eligibility. (It is also assumed in both calculations that all birthdays are equally likely. But if | |||
282 William Smiths registered in New Jersey in 2004. | they are not, then coincidences [https://www.maa.org/sites/default/files/M19526._Lawrence_Clevenson_and_William_Watkins.pdf are even more likely] ). | ||
For common names, this is a very real problem. For instance, the article reports that | |||
282 William Smiths were registered in New Jersey in 2004. As an example of how such coincidences can accumulate, the "One person, one vote" paper found that In 2012 and 2014, the Crosscheck program found nearly 240,000 voter registrations in Iowa that matched the name and birthdate of a record in some other state. | |||
Beyond these probabilistic worries, the article notes the mechanical issues integrating records across states and municipalities, with the challenges of missing data fields, transcription errors, and the like. | |||
Submitted by Bill Peterson | |||
==Birthday cake bacteria== | |||
Coincidentally, on the subject of birthdays (not voting!), Paul Alper sent a link to the following paper: | |||
:[http://www.ccsenet.org/journal/index.php/jfr/article/view/67217 Bacterial transfer associated with blowing out candles on a birthday cake]<br> | |||
:''Journal of Food Research'', Vol 6, No 4 (Aug 2017) | |||
The abstract is reproduced below: | |||
<blockquote> | |||
This study examined the potential spread of bacteria when blowing out candles on a birthday cake. Preliminary tests of blowing on nutrient agar indicated that bioaerosols in human breath expelled from the mouth may be a source of bacteria transferred to cake surfaces. To test aerosol transfer to cake, icing was spread evenly over foil then birthday candles were placed through the foil into a Styrofoam™ base. '''After consuming pizza''', test subjects were asked to extinguish the candles by blowing. Icing samples were sterilely recovered then surface plated, to determine the level of bacterial contamination. Blowing out the candles over the icing surface resulted in '''1400% more bacteria compared to icing not blown on'''. Due to the transfer of oral bacteria to icing by blowing out birthday candles, the transfer of bacteria and other microorganisms from the respiratory tract of a person blowing out candles to food consumed by others is likely. (Paul's emphasis) | |||
</blockquote> | |||
Later, in the methods section we read, "The experiment was replicated 3 times on separate days by 11 subjects yielding 33 observations per treatment | |||
(blow or no blow)." | |||
As Paul notes, a number of questions immediately suggest themselves, including the elephant in the party hat: What's pizza got to do with it? | |||
'''Discussion'''<br> | |||
What questions might you have based on the description above? To what extent are they dealt with in the [http://www.ccsenet.org/journal/index.php/jfr/article/view/67217/37099 full paper] (5 pages). | |||
==Surprise ER bills== | |||
[https://www.nytimes.com/2017/07/24/upshot/the-company-behind-many-surprise-emergency-room-bills.html The company behind many surprise emergency room bills]<br> | |||
Julie Creswell, Reed Abelson and Margot Sanger-Katz, "TheUpshot" blog, ''New York Times'', 24 July 2017 | |||
The article reports on a disturbing trend in medical billing. In some cases, patients receiving emergency room care at hospitals in networks covered by their insurance can still get large bills after being treated by out-of-network doctors. A [http://www.nber.org/papers/w23623 paper] by researchers at Yale University found that about 22% of ER visits overall in the US incur out-of-network charges. However, as explained in the ''NYT'' article, such experiences are more prevalent in hospitals that used a company named EmCare to manage their ER staffing. | |||
The article presents the following histogram, which gives a good example of a U-shaped distribution. | |||
<center> | |||
[[File:Emergency room.png|500px]] | |||
</center> | |||
==Rats, again!== | |||
[http://www.slate.com/articles/health_and_science/science/2017/07/de_blasio_s_team_consistently_misreads_the_stats_when_broadcasting_victories.html De Blasio wants to dramatically reduce NYC’s rat population. Don’t hold your breath.]<br> | |||
by Jonathan Auerbach, ''Slate'', 21 July 2017 | |||
New York City mayor Bill DiBlasio has announced a $32 million [http://www1.nyc.gov/office-of-the-mayor/news/472-17/de-blasio-administration-32-million-neighborhood-rat-reduction-plan#/0 plan to reduce the city's rat population]. In | |||
Chance News 102, we described a creative [https://www.causeweb.org/wiki/chance/index.php/Chance_News_102#Rats.21_NYC_capture-recapture_experiment capture-recapture experiment] that Auerbach implemented to estimate the size of that population. His results cast serious doubt on the folk wisdom that the city had more rates than people. | |||
Reducing the rat population still sounds like a good idea. However, in the ''Slate'' article, Auerbach questions the mayor's enthusiastic goals of a 70% reduction in select areas, which fits the pattern of other impressive-sounding claims about improvements in homicide rates, violent crime and pedestrian fatalities. He worries that by responding to extreme events, the apparent improvements may be due to the regression effect. The article alluded to Galton's famous data on heritability of height, where he found that sons of taller men tended to be above average in height, though not by quite as much as their fathers. (The ''BMJ'' has a good tutorial on [http://www.bmj.com/content/308/6942/1499 regression to the mean] that gives more detail on Galton's findings; a [http://www.bmj.com/content/309/6957/780 followup piece] gives further examples of the phenomenon.) | |||
The ''Slate'' article provides a number of data graphics to illustrate the problem in the context of the mayor's initiatives. For example. reproduced below is analysis of the Action Plan for Neighborhood Safety, implemented in 2014. The leftmost panel shows the 5-year trend in the violent crime in 13 neighborhoods targeted by the program from 2011 through 2015. The drop in 2015 was cited as evidence of the success of the program. | |||
<center> | |||
[[File:NYC crime Slate.png | 500px]] | |||
</center> | |||
But numbers of violent crimes fluctuate from year to year. The next two panels show what happened in the 13 highest crime developments from 2013 and 2012, respectively. Of course, the mayor's program was not in force in those year. Yet In both cases, the high-crime developments saw a drop in crime the following year that was comparable in magnitude that was observed in the neighborhoods targeted by the program. This demonstrates that the progress attributed to the program could simply be a statistical artifact. | |||
The article links to a paper entitled [http://mcadams.posc.mu.edu/econ/Friedman,%2520The%2520Regression%2520Fallacy.pdf Do old fallacies ever die?], in which Nobel Prize-winning economist Milton Friedman wrote, "I suspect that the regression fallacy is the most common fallacy in the statistical analysis of economic data." | |||
'''Discussion'''<br> | |||
The caption to the graphs shown above reads, "All three panels show increases from 2011 to the year in which the developments were selected, ''before finally reverting toward their 2011 levels by 2015'' (emphasis added). What do you make of this last comment? | |||
Submitted by Bill Peterson |
Latest revision as of 18:24, 7 August 2018
July 1, 2017 to August 20, 2017
Quotations
“In my darkest moods I follow what could be called the ‘Groucho principle’: because stories have gone through so many filters that encourage distortion and selection, the very fact that I am hearing a claim based on statistics is reason to disbelieve it.”
'Exaggerations' threaten public trust in science, says leading statistician, Guardian, 28 June 2017
"Aside from ruining the rest of the 21st century, the 2016 U.S. election inflicted spectacular collateral damage on the subject known as statistics."
"Indeed, the statistician David Freedman used to say that if the topic of regression comes up in a criminal or civil trial, the side that must explain regression to the jury will lose the case."
Forsooth
COLONEL [Buzz] ALDRIN: Infinity and beyond. (Laughter.)
THE PRESIDENT: This is infinity here. It could be infinity. We don’t really don’t know. But it could be. It has to be something -- but it could be infinity, right?
Okay. (Applause.)
Office of the White House Press Secretary, 30 June 2017.
Suggested by Mike Olinick
Do we have to?
Photo by Pete Schumer, taken in the elevator at his hotel.
Too much precision!
"By 2025, the [National Funeral Directors] Association is forecasting that 63.8 percent of the people who die in the United States will be cremated, and by 2035, 78.8 percent."
Submitted by Paul Alper
"Random" seat assignment
Ryanair's 'random' seat allocation not random — scientists
by John von Radowitz, Irish Independent, 30 June 2017
Ryanair is an Irish low-cost airline. This Wikipedia entry reports that they have repeatedly faced criticism for misleading advertising.
The Irish Independent article concerns a recent controversy. When customers book Ryanair flights, they can pay to make a seat selection or else opt for "random" seat assignment, which is free. Advertising on the airline's website says, "Can't stand the middle seat? Don't leave it to chance, take your pick from a choice of seats. Get up to 50pc off reserved seats with prices starting at £2."
But is it up to chance? In light of customer complaints, the BBC consumer affairs show Watchdog sought expert opinion from Oxford University. To test the claim, researchers had four groups of four passengers book travel on four separate flights, all under the random seating option. On every flight, all of the passengers got middle seats. The odds of this happening were estimated at about 1:540,000,000, much longer than the 1:45,000,000 odds of winning the UK National Lottery jackpot. The director of Oxford University's Statistical Consultancy, Dr. Jennifer Rogers, is quoted as saying, "This is a highly controversial topic and my analysis cast doubt on whether Ryanair's seat allocation can be purely random."
The article concludes with the following explanation from Ryanair, which may qualify as an extended Forsooth!
We haven't changed the random seat allocation policy.
The reason for more middle seats being allocated is that more and more passengers are taking our reserved seats (from just £2) and these passengers overwhelmingly prefer aisle and window seats which is why people who choose random (free of charge) seats are more likely to be allocated middle seats.
Some random seat passengers are confused by the appearance of empty seats beside them when they check-in up to four days prior to departure.
The reason they can't have these window or aisle seats is that these are more likely to be selected by reserved seat passengers, many of whom only check in 24 hours prior to departure.
Since our current load factor is 95pc, we have to keep these window and aisle seats free to facilitate those customers who are willing to pay (from £2) for them.
Submitted by Patrick O'Beirne
Discussion
- Emil Friedman wrote to say that the airline's explanation makes sense if we understand timing: the random assignment will be made after the reserved seats are fixed. Should the advertisement be reworded to reflect this?
- Should there be some statement of (historical) odds, as on the lottery web page? How do you think the Oxford team estimated the odds?
Sense and Sensibility and Statistics
From The Word Choices That Explain Why Jane Austen Endures
by Kathleen A. Flynn and Josh Katz, New York Times, 6 July 6, 2017.
Jane Austen' popularity has endured, and there may be something in her language that explains this. A principal components analysis of the words used by a large number books published from 1701 t0 1920 show that Austen's novels were unusual in her use of words related to time (always, fortnight and week) or emotion (awkward, decided, dislike, glad, sorry, suppose).
Jane Austen also uses a large number of intensifying words: quite, really, and very. These words are normally avoided by authors, but Austen uses them to develop a sense of irony. This fits in well with some non-statistical assessments.
Traditional literary approaches to Austen have long focused on this aspect of her work: “the incongruities between pretence and essence, between the large idea and the inadequate ego," as the critic Marvin Mudrick put it. A look at passages where words like very are used frequently often finds the stated meaning conceivably at odds with the real one, the exaggeration subtly inviting doubt."
A nice interactive plot shows the first two principal components and you can hover over individual data points to see the book title.
Submitted by Steve Simon
Followup
Charting Literary Greatness With Jane Austen
by Kathleen A. Flynn, New York Times "Times Insider", 15 July 2017
Flynn, one of the authors of the above article, describes on how the Austen project arose. Of the quantitative approach, she writes
To reduce words to data points is not to diminish the power of choosing the right ones and putting them in the right order; there’s magic in fiction’s ability to explore universal human truths. But a discipline like statistics can offer a new and amusing — even instructive — look at something as intuitive and artistic as a novel, and that’s exciting, too.
The aritcle provides links to other data-analytic studies of literature. For example, scholars at University of Nebraska-Lincoln have computed word frequencies for all of Austen's works, while the Stanford Literacy Project has applied "computational criticism" to a wide range of literary analyses.
Do the stats!
Statistical errors are often not due to mathematical errors
by Brian Zaharatos, letter to the editor, Chronicle of Higher Education, 11 July 2017
Zaharatos writes in response to a June 28 article from the Chronicle, entitled A new theory on how researchers can solve the reproducibility crisis: Do the math. While acknowledging some good points raised in the article, he notes that it also misinterprets several statistical concepts, as in this faulty description of significance:
[F]or a p-value of 0.05…a study’s finding will be deemed significant if researchers identify a 95-percent chance that it is genuine.
Regarding the headline, notes that people's difficulties interpreting statistical studies are not purely mathematical: "Mathematical skills are necessary for success in statistics, but they are far from sufficient. Statisticians — and researchers using statistics — ought to have a nuanced understanding of the concepts that mathematics helps them quantify."
Gerrymandering
Partisan gerrymandering has benefited the GOP, analysis shows
NBCnews.com, 25 June 2017
Republican electoral gains in 2010 gave the party substantial control over the redistricting that followed the 2010 Census. NBC reports on a statistical analysis by the Associated Press showing the effect in subsequent elections and the state and national levels. For example, in the 2016 election, they estimate that "Republicans won as many as 22 additional U.S. House seats over what would have been expected based on the average vote share in congressional districts across the country." The AP website reports that Texas had the largest gain, nearly 4 extra seats.
Gerrymandering is of course nothing new. The reference to Texas brought to mind a famous story, retold in a 2002 story from the Economist (How to rig an election, 25 April 2002): "[It] used to be said of the old Texas 6th (which was a road from Houston to Dallas), that you could kill most of the constituents by driving down the road with the car doors open."
However, the latest efforts are seen as particularly egregious. We read,
The AP’s analysis was based on an “efficiency gap” formula developed by University of Chicago law professor Nick Stephanopoulos and Eric McGhee, a researcher at the nonpartisan Public Policy Institute of California. Their mathematical model was cited last fall as “corroborative evidence” by a federal appeals court panel that struck down Wisconsin’s Assembly districts as an intentional partisan gerrymander in violation of Democratic voters’ rights to representation. The U.S. Supreme Court has agreed to hear an appeal.
Stephanopoulos and McGhee computed efficiency gaps for four decades of congressional and state House races starting in 1972, concluding the pro-Republican maps enacted after the 2010 Census resulted in “the most extreme gerrymanders in modern history.”
Here is how Stephanopoulos explained the efficiency gap idea in a 2014 article, Here's how we can end gerrymandering once and for all (New Republic, 2 July 2014):
The efficiency gap is simply the difference between the parties’ respective wasted votes in an election, divided by the total number of votes cast. Wasted votes are ballots that don’t contribute to victory for candidates, and they come in two forms: lost votes cast for candidates who are defeated, and surplus votes cast for winning candidates but in excess of what they needed to prevail. When a party gerrymanders a state, it tries to maximize the wasted votes for the opposing party while minimizing its own, thus producing a large efficiency gap. In a state with perfect partisan symmetry, both parties would have the same number of wasted votes.
NBC News also cites another analysis approach, introduced by the Princeton University Gerrymandering Project, launched this year by Sam Wang. The inaugural post includes this graphic
![Princeton gerrymander.png](/wiki/chance/images/thumb/f/fa/Princeton_gerrymander.png/450px-Princeton_gerrymander.png)
which shows how the last three election cycles have been unusual.
Wang has proposed three statistical tests for detecting gerrymandering. The Princeton Project site has a good description of how the tests work, and provides for interactive demonstrations.
The NBC News story summarized results from the above tests by saying,
...the extreme Republican advantages in some states were no fluke. The Republican edge in Michigan's state House districts had only a 1-in-16,000 probability of occurring by chance; in Wisconsin's Assembly districts, there was a mere 1-in-60,000 likelihood of it happening randomly...
These are p-values, which are often misinterpreted in news stories. (Chance News reader) Jeff Eiseman provided the following clarification.
If the relationship between the Democrat’s share of national votes and the Democrat’s share of sheets in Michigan were consistent with those before 2012, the chances that the Republicans would have achieved an edge as large or larger than they did in Michigan and Wisconsin would be only 1 out of 16,000 and 1 out of 60,000 respectively.
Submitted by Bill Peterson
More on middle-age mortality
Further criticism of social scientists and journalists jumping to conclusions based on mortality trends
by Andrew Gelman, "Statistical Modeling, Causal Inference, and Social Science" blog, 11 July 2017
In Deaths of despair from the last installment of Chance News, we described the Angus-Deaton work on middle-age mortality, including commentary by Gelman and others. In this latest post, Gelman includes links to some other articles suggested by his readers.
Gelman reiterates his support for the project, which he believes has raised important issues. But he points out that the omission age-adjustment was strange, since "reporting trends in mortality rates without age adjusting is like reporting trends in nominal prices without adjusting for the consumer price index."
Gelman and collaborators have produced a wealth of graphics related to the story, which are packaged in another post, Easier-to-download graphs of age-adjusted mortality trends by sex, ethnicity, and age group.
Greece prosecutes former chief statistician
Greece’s troubling prosecution of its former chief statistician: Why Congress and State should speak up
by Barry D. Nussbaum and Katherine K. Wallman, Huffington Post, 7 July 2017
The authors of this op-ed are present and past presidents of the American Statistical Association. They describe how the former chief statistician of Greece, Andreas Georgiou, is being prosecuted for his reporting on the Greek national debt. The issue is not his work, which has actually produced a more accurate accounting, but rather the revelation that the debt is larger than previously understood, which will require stricter austerity measures. Nussbaum and Wallman call for US government officials to stress the need for transparent statistical data, and more generally to condemn attacks on science.
In the US, statisticians are not being prosecuted for their analyses. However, the Greek saga calls to mind how climate scientists have been subjected to harrassment. Indeed, in February Representative Lamar Smith (R-Texas), chair of House Science Committee, convened hearings dubiously titled “Making EPA Great Again”.
In another recent story, the director of the US Census Bureau John Thompson announced his resignation this May. During an interview with NPR, Could a census without a leader spell trouble In 2020? (NPR "Code Switch", 15 July 2017), Thompson expressed confidence that work at the bureau would continue its work without disruption. Some observers have expressed concern because the 2020 census is fast approaching. Allocation of funds for many federal programs is dependent on census data. The counts are also have implications for Congressional redistricting. The Trump adminstration has so far been slow at filling top jobs.
Submitted by Bill Peterson
ASA statement on Georgiou case
ASA August Member News reports that Georgiou has been convicted by an appeals court in Greece. Earlier this month, Georgiou delivered a presentation on National Governments, Coerced Narratives, Creative Language, and Alternative Facts at the Joint Statistical Meetings. The ASA has issued the following statement:
We were disturbed to learn this morning of the conviction of Andreas Georgiou, the former president of the Hellenic Statistical Authority (ELSTAT), on charges of violation of duty by an appeals court in Greece. Dr. Georgiou was acquitted of these charges in December 2015; the decision to retry him amounts to double-jeopardy. As noted in our March 27 letter to Alexis Tsipras, Prime Minister of the Hellenic Republic, “There is widespread agreement among statistical authorities that Dr. Georgiou compiled Greek public finance statistics in accordance with EU rules and standards, with which Greece is obliged to comply, and that the charges against him are baseless.” We reiterate the call that: “all remaining charges against Dr. Georgiou and his former colleagues be dropped immediately. Under no circumstances can criminal sanctions for developing and presenting independent and objective statistical data in a credible way be justified."
Still running against Hillary?
Paul Alper sent a link to the following article.
- Trump can usually make it about a third of the way through an interview without mentioning Hillary Clinton
- by Philip Bump, Washington Post, 21 July 2017
Political observers have noted that President Trump thrives on campaign-style events, and that his communications regularly lapse into campaign rhetoric. The Post presents a quantitative analysis of how often Trump invokes the name of his 2016 opponent, Hillary Clinton, and other terms relating to the election. The following graphic from the article shows how many of Trump's 19 interviews to date have included such references, and how far into the interview the first reference appears.
Hillary Clinton's name is the "winner", coming up in 17 interviews, appearing on average less than 40% of the way through!
Moreover, the Post found that the two interviews without Hillary references (coded as appearing 100% of the way through) came early in his term, whereas she was mentioned almost immediately in his recent July 19 interview with the New York Times.
Similar graphs are constructed for the other terms.
Links to the interviews themselves are provided at the end of the article. Quoting the beginning of the transcript of that last interview from the Times:
TRUMP: Hi fellas, how you doing?
BAKER: Good. Good. How was your lunch [with Republican senators]?
TRUMP: It was good. We are very close. It’s a tough — you know, health care. Look, Hillary Clinton worked eight years in the White House with her husband as president and having majorities and couldn’t get it done. Smart people, tough people — couldn’t get it done. Obama worked so hard... (emphasis added)
As the Post reported, we can see that he got though just 20 words before "Hillary." The caption for the second graphic says that the figures are based on "character analysis." This is not explicitly defined in the article, but the preceding suggests that it means word counts, calculating the number of words he spoke before the first "Hillary" divided by his total words in the interview.
Discussion
- From the first graphic, it appears that the terms that came up in many interviews also tended to come up early. Is this surprising? Can you think of another way to reveal this information?
- From the second graphic, consider distribution for the point at which "Hillary" first occurs. Does this square with headline? What is happening here?
Voter fraud and the birthday problem
This anti-voter-fraud program gets it wrong over 99 percent of the time. The GOP wants to take it nationwide.
by Christopher Ingraham, "Wonkblog", Washington Post, 20 July 2017
Throughout the 2016 campaign, candidate Trump warned that the election was being rigged. After the election, he complained that Hillary Clinton would not have won the popular vote without millions of illegal ballots. Such claims have cast a shadow over the recently convened Presidential Advisory Commission on Election Integrity, despite Vice President Pence's assurances that "this commission has no preconceived notions."
Kansas Secretary of State Kris Kobach is the vice-chairman of the commission. As described in the Post, he has advocated national implementation of Kansas program called Interstate Voter Registration Crosscheck, which seeks to identify people who may be voting in more than one location. Other states are invited to send their voter registration lists to Kansas, and Crosscheck will search for voters with matching first name, last name and date of birth. The idea is to remove duplicate registrations, though it is not exactly clear how this would be implemented.
Critics of the program argue that it creates many "false positives"; that is, different people with matching names and birthdays. From the article we read:
A statistical analysis of the program (One person, one vote: estimating the prevalence of double voting in U.S. presidential elections ) published earlier this year by researchers at Stanford, Harvard, University of Pennsylvania and Microsoft, for instance, found that Crosscheck “would eliminate about 200 registrations used to cast legitimate votes for every one registration used to cast a double vote.”
Ostensibly, Crosscheck appears to assume that a match on name and birthdate indicates a duplicate registration by the same person. Calculating the chance of such a match among different people is an application of a famous problem from probability. The Post cites a 2007 paper, Seeing double voting: An extension of the birthday problem. It is well-known that among 23 randomly selected birthdays (month and day), there is a better than even chance that at least two will match. With 180 people, the article reports that there is a 50% chance that two will match birthdates (month, day and year). There is a built-in function in R called qbirthday that implements such calculations. For example:
> qbirthday(prob=0.5, classes=365, coincident=2)
[1] 23
> qbirthday(prob=0.5, classes=365*65, coincident=2)
[1] 182
For the second calculation we assume that an individual has 65 years of voting eligibility. (It is also assumed in both calculations that all birthdays are equally likely. But if they are not, then coincidences are even more likely ).
For common names, this is a very real problem. For instance, the article reports that 282 William Smiths were registered in New Jersey in 2004. As an example of how such coincidences can accumulate, the "One person, one vote" paper found that In 2012 and 2014, the Crosscheck program found nearly 240,000 voter registrations in Iowa that matched the name and birthdate of a record in some other state.
Beyond these probabilistic worries, the article notes the mechanical issues integrating records across states and municipalities, with the challenges of missing data fields, transcription errors, and the like.
Submitted by Bill Peterson
Birthday cake bacteria
Coincidentally, on the subject of birthdays (not voting!), Paul Alper sent a link to the following paper:
- Bacterial transfer associated with blowing out candles on a birthday cake
- Journal of Food Research, Vol 6, No 4 (Aug 2017)
The abstract is reproduced below:
This study examined the potential spread of bacteria when blowing out candles on a birthday cake. Preliminary tests of blowing on nutrient agar indicated that bioaerosols in human breath expelled from the mouth may be a source of bacteria transferred to cake surfaces. To test aerosol transfer to cake, icing was spread evenly over foil then birthday candles were placed through the foil into a Styrofoam™ base. After consuming pizza, test subjects were asked to extinguish the candles by blowing. Icing samples were sterilely recovered then surface plated, to determine the level of bacterial contamination. Blowing out the candles over the icing surface resulted in 1400% more bacteria compared to icing not blown on. Due to the transfer of oral bacteria to icing by blowing out birthday candles, the transfer of bacteria and other microorganisms from the respiratory tract of a person blowing out candles to food consumed by others is likely. (Paul's emphasis)
Later, in the methods section we read, "The experiment was replicated 3 times on separate days by 11 subjects yielding 33 observations per treatment (blow or no blow)."
As Paul notes, a number of questions immediately suggest themselves, including the elephant in the party hat: What's pizza got to do with it?
Discussion
What questions might you have based on the description above? To what extent are they dealt with in the full paper (5 pages).
Surprise ER bills
The company behind many surprise emergency room bills
Julie Creswell, Reed Abelson and Margot Sanger-Katz, "TheUpshot" blog, New York Times, 24 July 2017
The article reports on a disturbing trend in medical billing. In some cases, patients receiving emergency room care at hospitals in networks covered by their insurance can still get large bills after being treated by out-of-network doctors. A paper by researchers at Yale University found that about 22% of ER visits overall in the US incur out-of-network charges. However, as explained in the NYT article, such experiences are more prevalent in hospitals that used a company named EmCare to manage their ER staffing.
The article presents the following histogram, which gives a good example of a U-shaped distribution.
Rats, again!
De Blasio wants to dramatically reduce NYC’s rat population. Don’t hold your breath.
by Jonathan Auerbach, Slate, 21 July 2017
New York City mayor Bill DiBlasio has announced a $32 million plan to reduce the city's rat population. In Chance News 102, we described a creative capture-recapture experiment that Auerbach implemented to estimate the size of that population. His results cast serious doubt on the folk wisdom that the city had more rates than people.
Reducing the rat population still sounds like a good idea. However, in the Slate article, Auerbach questions the mayor's enthusiastic goals of a 70% reduction in select areas, which fits the pattern of other impressive-sounding claims about improvements in homicide rates, violent crime and pedestrian fatalities. He worries that by responding to extreme events, the apparent improvements may be due to the regression effect. The article alluded to Galton's famous data on heritability of height, where he found that sons of taller men tended to be above average in height, though not by quite as much as their fathers. (The BMJ has a good tutorial on regression to the mean that gives more detail on Galton's findings; a followup piece gives further examples of the phenomenon.)
The Slate article provides a number of data graphics to illustrate the problem in the context of the mayor's initiatives. For example. reproduced below is analysis of the Action Plan for Neighborhood Safety, implemented in 2014. The leftmost panel shows the 5-year trend in the violent crime in 13 neighborhoods targeted by the program from 2011 through 2015. The drop in 2015 was cited as evidence of the success of the program.
But numbers of violent crimes fluctuate from year to year. The next two panels show what happened in the 13 highest crime developments from 2013 and 2012, respectively. Of course, the mayor's program was not in force in those year. Yet In both cases, the high-crime developments saw a drop in crime the following year that was comparable in magnitude that was observed in the neighborhoods targeted by the program. This demonstrates that the progress attributed to the program could simply be a statistical artifact.
The article links to a paper entitled Do old fallacies ever die?, in which Nobel Prize-winning economist Milton Friedman wrote, "I suspect that the regression fallacy is the most common fallacy in the statistical analysis of economic data."
Discussion
The caption to the graphs shown above reads, "All three panels show increases from 2011 to the year in which the developments were selected, before finally reverting toward their 2011 levels by 2015 (emphasis added). What do you make of this last comment?
Submitted by Bill Peterson