![]() Within this data is an example for every test we teach in Introduction to Statistics. Yes, I know that actual transformation is much more involved than this, but it is a simple way to introduce the topic. ![]() This could be a simple way to introduce the topic. ![]() I don't know, but I can assume that we've moved on from The Bucket Method, which was the best they had at the time.ĭata transformations: These can be tricky things to explain to the novice, like transformations to make data less skewed. ![]() I assume that it involves sharks with lasers on their heads. We can accurately measure water temperature now. This is the story of graduate student Duo Chan and his efforts to make his archival data as accurate as possible. That Air Force PDF uncovered by the researchers' friends is also a good example of why you need to maintain a codebook and extensive notes on how you treat your data. I also like how they could tell that something was going on in their data but didn't know WHY something was going on in their data. But now we have this opportunity for older data, right, so we can better understand the larger cycle of global water temperatures. We have only been tracking weather for 100 years. Which is not ideal, but you can correct for it.Īrchival data: Sometimes, the data you need already exists somewhere. Systematic bias: The data were all flawed in the same way as it was transcribed without any data to the right of the decimal point. Below, I will highlight some of the teaching items. I think this story might be a little beyond Intro Stats but it tells the story of a) messy, real archival data used to inform global climate change and b) introduces the idea of data transformations. Instead, she told the story of how the researchers discovered and corrected for their flawed ocean water temperature data. Reporter Rebecca Hersher didn't discuss the entire research paper. Here is the original research, published in Nature. ^Go to the 3 minute mark to see the bucket-boat-water-temperature technique in action In two very high profile spying cases (Aldrich Ames and Jerry Chun Shing Lee) the spies were uncovered not by an algorithm but by human analysts who noticed odd behaviors and acted on those observations. The bigger narrative in this piece has to do with how various government agencies attempt to use algorithms to uncover likely spies within their ranks. An algorithm can't know this.Īlgorithms still can't beat human insights (n = 2): But a new parent may also work very early hours so they can leave early to accommodate their kid's school schedule, or an employee might stay late routinely because they have a regular, early PT appointment. Similarly, people who are engaging in corporate espionage tend to show up before everyone else or stay after everyone else. There are a lot of those! If the federal government flagged everyone who met that description, they would flag many, many non-spies. Who are spies? White men who work for the US government and speak Russian. "By learning from the past, algorithms are doomed to repeat the past" (Another great one-liner!) Which isn't to say that there aren't qualified white dudes, but you will miss out on great women and POC. If that is part of your algorithm, you'll keep promoting only white dudes. "The feedback loop that reinforces lucky people's luck." Historically, who is likely to get promoted at large, successful organizations? White dudes. Sometimes, I feel like I'm just waving my hands when I try to explain this very, very important piece of regression but this report describes the prediction side of regression succinctly. So, regression, right? Fancy regression, but that one line can take this fancy talk of algorithms and make it more applicable to your students. "Algorithms use past patterns of success to predict the future." They talk about regression without ever saying "regression": It clocks in at just under five minutes, perfect as a discussion prompt or quick introduction to the topic. This NPR story by Dina Temple-Raston is a great primer for All The Ethical Issues Related To Algorithms, accessible to non- or novice-statisticians. There are many great one-liners in this little five-minute review of algorithms. ^This quote from this NPR story made me punch the air in my little Subaru after dropping my kid off to school. "(Algorithms) are used so heavily, they don't just predict the future, they are the future." -Cathy O'Neil
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |