UPDATE: 2020 Data: https://www.catalyst.org/knowledge/women-government When I explain chi-square at a conceptual, no-software, no-formula level, I use the example of gender distribution within the US Senate. There are 100 Senators, so the raw observed data count is the same as the observed data expressed via proportions. I think it makes it easier for junior statisticians to wrap their brains around chi-square.  I  usually start with an Goodness-of-Fit (or, as I like to call them, "One-sies chi-squares").For this example, I divide senators into two groups: men and women. And what do you get?  For the 115th Congress, there are 23 women and 77 men . There is your observed data, both as a raw count or as a proportion. What is your expected data? A 50/50 breakdown...which would also be 50 men and 50 women. Without doing the actual analysis, it is pretty safe to assume that, due to the great difference between expected and observed values, your chi-square Goodness o...

https://pudding.cool/2018/01/chyrons/ This example doesn't demonstrate a specific statistical test. Instead, it demonstrate how data can be used to answer a hotly contested question: Are certain media outlets biased? How can we answer this? Charlie Smart, working for The Pudding, addressed this question via content analysis. Here is how he did it: And here are some of their findings: Yes, Fox News was talking about the Clintons a lot. While over at MSNBC, they discussed the investigation into Russia and the 2016 elections ore frequently. While kneeling during the anthem was featured on all networks, it was featured most frequently on Fox And context matters. What words are associated with "dossier"? How do the different networks contextualize President Trump's tweets? Another reason I like this example: It points out the trends for the three big networks. So, we aren't a bunch of Marxist professors ragging on FOX, and we ar...

As I've discussed in this space before, I think that it is just as important to show our students how to use statistics in real life as it is to show our students how to conduct an ANOVA. The "You Draw It" series from the New York Times provides an interactive, personalized example of using data to prove a point and challenge assumptions. Essentially, this series asks you to predict data trends for various social issues. Then it shows you how the data actually looks. So far, there are three of these features: 1) one that challenges assumptions about Obama's performance as president, 2) one that illustrates the impact of SES on college attendance, and 3) one that illustrates just how bad the opioid crisis has become in our country. Obama Legacy Data This "You Draw It" asks you to predict Obama's performance on a number of measures of success. Below, the dotted yellow line represents my estimate of the national debt under Obama. The blue line shows t...

This article from The Washington Post summarizes research published in the Journal of the American Medical Association . Both are simple, short articles that show how you can use regression to make an argument. Here, the authors use regression to demonstrate the paucity of funding and publications for research studying gun-related deaths. A regression line was generated to predict how much money was spent studying common causes of death in the US. Visually, we can see that deaths by firearms aren't receiving funding proportional to the number of deaths they cause. See the graph below. How to use in class: 1) How is funding meted out by our government to better understand the problems that plague our country? Well, it isn't being given to researchers studying gun violence because of the Dickey Amendment . I grew up in a very hunting friendly/gun-friendly part of Pennsylvania. I've been to the shooting range. And it upsets me that we can't better understand and stu...

Bahahaha. A very funny example of conflict of interest, as this satirical study was sponsored by Red Robin.  Click through to the original content to rea d how the study replaced "I Voted" stickers with " thick Red Robin Gourmet Cheeseburger complete with pickle relish, tomatoes, onions, lettuce, mayonnaise, and their choice of cheese". http://creative.theonion.com/ads/onion-ring/article/study-giving-away-ldquoi-votedrdquo-burger-instead-of-sticker-would-increase-voter-turnout-by-80

CNN posted a whole mess of exit polling data that illustrates how different demographics voted last night. And through my "I teach too many stats classes" lense, I see many examples of chi-square. I think they work at a conceptual level to clearly illustrate how chi-square looks at people falling into different categories, then measures whether the distribution of people is by chance or not. If you actually wanted to test these using chi square, I would suggest you should problem delete the other/no answer column (or else they will all come out as statistical significant, I would wager). EDIT (11/14/16); Daniel Findley made of video demonstrating how to use Excel to conduct chi-square tests on the marital status data. Check it out here .

This interactive tool from Pew research illustrates left and right skew as well as median and longitudinal data. The x-axis indicates how politically consistent (as determined by a survey of political issues) self-identified republicans and democrats are across time. Press the button and you can animate data, or cut up the data so you only see one party or only the most politically active Americans. http://www.people-press.org/2014/06/12/section-1-growing-ideological-consistency/#interactive The data for both political part goes from being normally distributed in 1994 to skewed by 2014. And you can watch what happens to the median as the political winds change (and perhaps remind your students as to why mean would be the less desirable measure of central tendency for this example). I think it is interesting to see the relative unity in political thought (as demonstrated by more Republicans and Democrats indicating mixed political opinions) in the wake of 9/11 but more politicall...

This headline is a good example of a) journalists misrepresenting statistics as well as b) confidence intervals/margin of error more broadly. See the headline below: In actuality, Bernie didn't exactly take the lead over Hillary Clinton. Instead, a Quinnipiac poll showed that 41% of likely Democratic primary voters in Iowa indicated that they would vote for Sanders, while 40% reported that they would vote for Clinton. If you go to the original Quinnipiac poll , you can read that the actual data has a margin of error of +/- 3.4%, which means that the candidates are running neck and neck. Which, I think, would have still been a compelling headline.  I used this as an example just last week to explain applied confidence intervals. I also used this as a round-about way of explaining how confidence intervals are now being used as an alternative/compliment to p -values. 

Aschwanden (for fivethirtyeight.com) did an extensive piece that summarizes that data/p-hacking/what's wrong with statistical significance crisis in statistics. There is a focus on the social sciences, including some quotes from Brian Nosek regarding his replication work. The report also draws attention to  Retraction Watch  and Center for Open Science as well as retractions of findings (as an indicator of fraud and data misuse). The article also describes our funny bias of sticking to early, big research findings even after those research findings are disproved (example used here is the breakfast eating:weight loss relationship). The whole article could be used for a statistics or research methods class. I do think that the p-hacking interactive tool found in this report could be especially useful illustration of How to Lie with Statistics. The "Hack your way to scientific glory" interactive piece demonstrates that if you fool around enough with your operationalized...

Pew recently revisited the question of how survey modality can influence survey responses.  In particular, this survey used both web and telephone based surveys to ask participants about their attitudes towards politicians, perceptions of discrimination, and their satisfaction with life. As summarized in the article, the big differences are: "1)  People expressed more negative views of politicians in Web surveys than in phone surveys."  "2)  People who took phone surveys were more likely than those who took Web surveys to say that certain groups of people – such as gays and lesbians, Hispanics, and blacks – faced “a lot” of discrimination ."  "3)  People were more likely to say they are happy with their family and social life when asked by a person over the phone than when answering questions on the Web ."     The social psychologist in me likes this as an example of the Social Desirability Bias. When spea...

http://www.washingtonpost.com/blogs/the-fix/wp/2014/09/29/ believe-it-or-not-some-members-of-congress-are-accountable-to-voters/ Philip Bump (writing for The Washington Post) illustrates the linear relationship between a U.S. House of Representative Representative's politics and their home district's politics. Yes, this is entirely intuitive. However, it is still a nice example of correlations/linear relationships for the reasons described below. Points for class discussion: 1) How do they go about calculating this correlation? What are the two quantitative variables that have been selected? Via legislative rankings (from the National Journal) on the y-axis and voting patterns from the House member's home district on the x-axis. 2) Several outliers' (perhaps not mathematical outliers, but instances of Representative vs. District mismatch ) careers are highlighted within the news story in order to explain why they don't align as closely with their distric...

This radio story  by Mara Liasson (reporting for NPR) discusses the surprising primary loss of former Republican House Majority Leader Eric Cantor. It was surprising because internal polling conducted by Cantor's team gave him an easy win, but he lost out to a Tea Party favorite, David Brat. The story goes on to describe why it is becoming increasingly difficult to conduct accurate voter polling via telephone and the internet. Some specific points from this story that teach students about sampling techniques: 1) Sample versus population: One limitation of polling data is the fact that many telephone call-based sampling techniques include landlines and ignore the growing population of people who only have cell phones. 2) Response rates for political polling are on a decline, making the validity of the available sample shrink. 3) Robocalls, while less expensive, have no way of validating that an actual registered voter is responding to the questions. Additionally, restrictio...

Robinson & Feltus, 2014 There appears to be a connection between political affiliation, likelihood to vote, and preferred adult beverage. If you lean right and drink Cabernet Savignon, you are more likely to vote than one who enjoys "any malt liquor" and leans left.  This Washington Post story summarizes data analysis performed by the  National Media Research Planning and Placement . NMRPP got their data from market research firm Scarborough . There is also a video embedded in the Washington Post story that summarizes the main findings. I think this is a good example of illustrating data as well as data mining pre-existing data sets for interesting trends. And beer.

This scale , created by Haidt and Wilson, predicts your political leanings based upon seemingly unrelated questions. Screen grab from time.com You can use this in a classroom to 1) demonstrate interactive, Likert-type scales, 2) face validity (or lack there of). I think this would be 3) useful for a psychometrics class to discuss scale building. Finally, the update at the end of the article mentions 4) both the n-size and the correlation coefficient for their reliability study, allowing you discuss those concepts with students. For more about this research, try yourmorals.org

Best. Awful. Headline. Ever. T his headline makes it sound like many young women support the sexting, bad-decision-making, former NY representative Anthony Weiner. However, if one takes a moment to read the article, one will learn that the "young women" sampled were recruited from SeekingArrangement.com. A website for women looking for sugar daddies. If you want your brain to further explode, read through the comments section for the article. Everyone is reacting to the headline. Very few people actually read through the article themselves...which provides further anecdotal evidence that most folks can't tell good data from bad (and that part of our job as statistics instructors, in my opinion, is to ameliorate this problem).

Prior to the 2012 presidential election, Pennsylvania attempted to enact one of the toughest voter ID laws in the nation. This law has been kicked up to the courts to examine its legality. One reason that so many people protested the law was because it would make it more difficult for the elderly and the poor to vote (as it would be more difficult for them to obtain the ID required). Here is an NPR story that gives a bit of background on the law and the case in court.   Also, for giggles and grins, here is Jon Stewart's more amusing explanation of the law and why it was struck down prior to the election, including video footage of a PA legislature flat-out stating that the Voter ID law would allow Romney to win the 2012 election. In order to support/raise questions about the impact of the law on the ability to vote, statisticians have been brought in on both sides in order to estimate exactly how disenfranchising this law will be. Essentially, the debate in court centers a...

One way of teaching about margin of error/confidence intervals is via political polling data. From  mvbarer.blogspot.com Here is a good site that has a break down of polling data taken in September 2012 for the 2012 US presidential election. I like this example because it draws on data from several well-reputed polling sites, includes their point estimates of the mean and their margin of errors. This allows for several good examples: a) the point estimates for the various polling organization all differ slightly (illustrating sampling error), b) the margin of errors  are provided, and c) it can be used to demonstrate how CIs can overlap, hence, muddying our ability to predict outcomes from point estimates of the mean. I tend to follow the previous example with this gorgeous polling data from Mullenberg College : This is how sampling is done, son! While stats teachers frequently discuss error reduction via big n , Mullenberg takes it a step further by o...