Net2Vic: Effective Data Analysis for Everyone (October 23, 2014)

Editor's Notes

• #2: Data is ubiquitous in our lives and work places, and more data can be easily collected. And those data can, if analyzed correctly, provide insights that lead to better decision making. Martin Monkman will present the key ideas for effective and meaningful data analysis, including sources of existing data, things to think about when collecting new data, analyzing results, and effective presentation and reporting of the data.
• #4: As I started to think about this, I realized that this is not a topic for a single hour – one could spend a lifetime researching, thinking, and practicing this topic. I have been actively engaged in this topic – in one capacity or another – for (gads) 30 years. And I’m still learning! To use the metaphor of this library, we’ll visit a couple of the stacks and pull of four books. There’s a wealth of information on the topics we’ll skim … some links to further reading are in the notes. =-=-=-=-=-=-=-= Image: Cincinnati-Old Main Library that was demolished in 1955, Source: https://www.flickr.com/photos/80868612@N03/11983800566/in/photolist-aiP4WA-aSvCV2-bu7Ctj-3AEWz-aSv5ZP-bjtFJV-bhkodF-bhkoBD-jfY7p9-bujoxS-bhknQt-4VJnag-4TZ8fJ-5Av8e4-5kfMPH-59oHHj-5GzcXQ-5CowXv Creative Commons
• #5: Why should you care about data analysis? Because you want to ensure that your limited resources – if you’re running a company, a non-profit, a government program, a household, or a car – are being used in their most effective way possible. To get your car to go 140 km/hour, all of the car’s systems need to be running at their optimal levels. The speedometer isn’t the only dial on the dashboard, and the mechanics we rely on have other diagnostic tools at their disposal. And there are things outside the car that we consider when operating it – an extreme case would be the choices that are made about racing car tires, which are weather and temperature dependent. Let’s imagine for a moment a coffee shop. And you’re the manager! The volume of customers ebbs and flows through the day – a rush first thing in the morning, and then a slow period in the middle of the afternoon before the evening crowd shows up. And there might also be fluctuations through the week; Monday might be busier than Sunday. You don’t want to have the same number of staff on at all times … it would be much more efficient to have more staff on at peak times, and fewer when the shop is less busy. Quantifying those ebbs and flows can help you make more efficient decisions about staffing your coffee shop. So where do we start?
• #6: So today I want to touch on four ideas for you that I hope will get you started. Image source: https://www.flickr.com/photos/duncan/3782076384 Creative commons
• #7: Five Ws - who, what, why, when, where
• #9: So today I want to touch on four ideas for you that I hope will get you started. Image source: https://www.flickr.com/photos/holeymoon/2098042797/in/photostream/ Creative commons
• #10: Five Ws - who, what, why, when, where
• #12: Administrative sources – this includes things such as registrations for programs, or transaction data. Much of what is part of “big data” comes from these sorts of administrative records. They are not collected for research purposes, but for the administration of business operations. Some examples: the City of Victoria’s parking meter data – provides insights into utilization across the system, and was used to modify the rates that are charged in different parts of the city grocery stores scan bar codes to keep track of their inventory, but these data points provide information about shopping patterns (what items are being purchased together, the time of day that certain items are selling – important for deli stock!) Or you can collect your own data, through a survey. Surveys – they sound easy, but they are fraught with peril!
• #13: Who is your population? Who should you be contacting? Are you considering households or individuals? How will you ensure that your respondents are representative of the entire population of interest? How are you going to reach them? - Web surveys sound easy
• #14: What NOT to do. The Johnson Street bridge, a.k.a. The Blue Bridge. The City of Victoria is in the process of replacing the bridge, and early on in the planning process had a “public consultation”. One element of the consultation process was what was described as a survey of the residents of the City, to gauge their preferences for three different bridge designs. The question was framed as “Do you like Design One, Design B, or Design Three”? There was no “None of the above”, nor was there a “Why don’t we just repair the existing bridge?” As well, there was no option for “Make sure it includes designated bike and pedestrian lanes”, and there wasn’t a “If there’s no rail line on the bridge, what does that mean for the idea of light rail commuter traffic into the downtown?” Another point: the questions were posed on an open website. THIS IS NOT A SURVEY. There were no controls over who could respond (i.e. the “population” was anyone on Earth with a web connection), nor were there controls over how many times one could respond (ballot box stuffing!) I took delight in responding to the survey three times, one for each design, and in the “Do you have any comments?” box, pointing out that a) I was responding multiple times to the survey and b) I don’t live in the City of Victoria. POST-PRESENTATION COMMENTARY: COMMENT #1: My lessons re: the Blue Bridge example -- 1. avoid explicit political opinions, and 2. when using complicated examples like this make sure you've memorized all the details! The Blue Bridge online survey I referenced ran in September 2009, at a time when Victoria City Council had not yet explicitly voted to move forward with replacement. Here's the summary from the Citizen Advisory Committee: http://www.johnsonstreetbridge.com/wp-content/uploads/2009/09/consulation_Backgrounder-Sept-24-09.pdf A few months later a much more reliable and robust Ipsos Reid survey was commissioned by the City (the results were tabled in May 2010). This survey included a variety of options, including rehabilitation (http://www.johnsonstreetbridge.com/wp-content/uploads/2010/05/JSB-Residential-Presentation.pdf) It was a full year after the online “survey” that the referendum on the bridge repair / replacement was held (November 2010; http://www.johnsonstreetbridge.com/public-involvement/get-informed-provide-input/). Even as late as November 2010, the notion of repair was still viewed as an option (see this article from Focus: http://focusonline.ca/?q=node/93). So I stand by my comment that other options beyond the three design choices could have been included in the initial questions. COMMENT #2: It was suggested that tracking IP addresses of the responses would be a way to limit ballot box stuffing. While this does could be used, this approach won’t get to a way to have a reliable sample of the defined population. It doesn’t address: the issue that a single respondent could use multiple IP addresses, it would limit the opportunity for multiple individuals in the same household to legitimately submit responses And finally, there is no reliable way to handle multiple responses from a single IP. Do you accept the first? The last? The average of them all? Image source: https://www.flickr.com/photos/wolfnowl/4976067226 Marcia and Mike Nelson Pedde Creative Commons
• #15: Data Analysis Image source: https://www.flickr.com/photos/holeymoon/1813639142 Creative commons
• #16: Here’s everything that they teach you in statistics class. There are a lot of different statistical tests to measure the significant of differences, and Greek names for distributions, and some really elaborate ways to measure relationships … And although the title of the presentation is Data Analysis for Everyone, this is going to be a one-slide section. Although I will point out that relationships – correlations – can be spurious. http://www.tylervigen.com/ Because when it comes to a presentation, nothing beats some good graphics. Which is Idea #4 …
• #17: Here’s some online courses for those interested in digging deeper into statistical analysis. The google course is a great introduction, which leans to the practical applications and away from detailed methodological issues. The others are more “academic” in inclination.
• #18: Communicating your analysis. There’s a whole dimension to writing about numbers that we won’t touch on – but let’s look at data visualization, the kind of communication that works for a presentation. Image source: https://www.flickr.com/photos/chrisinplymouth/7912514318/in/photolist-bewNF-abDZkN-7GXehN-89dz9F-6vP8ua-6urLeU-c8g4oj-6BCsT1-87zdmN-7R7FyL-aXen3-6NQvnd-cojYW1-bNQPLB-cEHfdh-dWn5PZ-amh4D3-7FcJ48-6BLKFB-6svVna-6sQUjk-6zbFq7-6Mk98X-6MSd2Q-7FWVn9-6iQgGQ-eksPZa-6BmTjJ-c9UW2f-c8g4QN-6FwgzD-6BPpzG-cR2sDo-d4cGQE-nGpLo4-pdVU4-4JTPJc-cKT99w-7V2djG-dWsuaS-9q1goR-czpP7w-7haSkT-a3Fw5r-aD4nkE-9yHvzR-8mTDhz-5bg7PR-6YTtnv-cpdv7Q Creative commons
• #19: Charts summarize data to make it more understandable That summary is often to compare different sets of data Exploratory Data Analysis - Graphics as a way to understand the data - Tukey, Cleveland Narrative - “data journalism” - telling a story with the data - answering the 5 W questions - Alberto Cairo
• #21: http://en.wikipedia.org/wiki/Anscombe%27s_quartet 1973, Francis Anscombe These four data sets of 11 x-y pairs have virtually identical summary statistics.
• #22: When we look at the four series in Anscombe’s Quartet, we can see that in spite of their identical summary statistics they are VERY different. Plotted in R http://www.r-bloggers.com/the-visual-difference-%e2%80%93-r-and-anscombe%e2%80%99s-quartet/ http://stat.ethz.ch/R-manual/R-patched/library/datasets/html/anscombe.html http://www.inside-r.org/node/36103
• #23: Minard’s (1869) famous map / data visualization shows Napoleon’s march to Moscow in the winter of 1812. Over 300,000 French troops set off – the thick brown line on the left side of the map. The brown line is much thinner by the time they got to Moscow ... And the black line, showing the size of the army on the return journey, is thinner still. The return trip: only 27,000 remained after the crossing of the Berezina River An in-depth analysis of the map is found in Edward Tufte’s “The Visual Display of Quantitative Information” and here: http://www.csiss.org/classics/content/58 Image: http://news.bbc.co.uk/2/hi/in_pictures/8206064.stm
• #24: Population pyramid’s are one of the demographer’s greatest tools, and allow us to quickly assess the population of a region. Here, the age and gender make up of the Capital region is compared to the population of British Columbia. It’s easy to see that the Capital region has a smaller proportion of children, roughly the same proportion of people in their 20s, fewer in their 30s – 50s, and more in the 60+ ages. Link: http://population-pyramid.herokuapp.com/#/Capital%252C%2520BC/British%2520Columbia http://bcstats.gov.bc.ca/StatisticsBySubject/ExperimentalDataTools.aspx
• #25: Even more dramatic is the population of Qualicum Beach – a town where 70 year olds feel like teenagers!
• #26: Tool for data analysis and data visualization: R FROM http://cran.r-project.org/ R is ‘GNU S’, a freely available language and environment for statistical computing and graphics which provides a wide variety of statistical and graphical techniques: linear and nonlinear modelling, statistical tests, time series analysis, classification, clustering, etc. Plus graphics, dynamic web display of data and charts, etc
• #27: This quick plot shows the population of the region by the 13 municipalities, and the unincorporated areas of the western communities and the Saanich peninsula and the Gulf Islands.
• #28: This revision groups the areas by the “core” municipalities of Victoria, Saanich, Oak Bay, and Esquimalt, the Saanich Peninsula & the Gulf Islands, and the Western Communities.* *A member of the audience pointed out that View Royal is now officially part of “the core”.
• #29: This chart shows the population of the three parts of the CRD since 1996
• #30: The most dramatic factor in the growth in the CRD in the past 20 years has been the fact that most of the growth in the past decade has been in the Western communities – in fact, in the past 3 years, the population of the core and Saanich peninsula / Gulf islands has fallen.
• #31: There are 68 Census Tracts in the Victoria Census Metropolitan Area (basically, the urbanized areas of the CRD). Part A - Short definition: Area that is small and relatively stable. Census tracts usually have a population between 2,500 and 8,000 persons*. They are located in census metropolitan areas and in census agglomerations that have a core population of 50,000 or more. *average of 4,000 See: Statistics Canada http://www12.statcan.gc.ca/census-recensement/2011/ref/dict/geo013-eng.cfm
• #32: There are 68 Census Tracts in the Victoria Census Metropolitan Area (basically, the urbanized areas of the CRD). Part A - Short definition: Area that is small and relatively stable. Census tracts usually have a population between 2,500 and 8,000 persons*. They are located in census metropolitan areas and in census agglomerations that have a core population of 50,000 or more. *average of 4,000 See: Statistics Canada http://www12.statcan.gc.ca/census-recensement/2011/ref/dict/geo013-eng.cfm
• #33: Census Tract Profiles http://www12.statcan.gc.ca/census-recensement/2006/dp-pd/prof/92-597/index.cfm?Lang=E For 0003.01 http://www12.statcan.gc.ca/census-recensement/2006/dp-pd/prof/92-597/P3.cfm?Lang=E&CTuid=9350003.01
• #34: Another example of R used for data visualization. Dialect map http://spark.rstudio.com/jkatz/SurveyMaps/ Carbonated beverage = Q105
• #36: Image: http://s684.photobucket.com/user/TexasChief/media/pie-i-have-eaten-chart.jpg.html What’s the point of a pie chart? To compare the relationship of the parts out of a whole Pie charts are actually bad at this Save the pies for desert (Stephen Few) Pie is delicious but not nutritious (Peter Flom) In general, circular depictions are discouraged We have a hard time discerning differences in size and angle No fixed reference for comparison Different colour slices can deceive our eyes, too – research has shown that “emphasis” can be judged as size as well
• #37: It’s a lot easier to see that the five categories are different sizes in the bar chart.
• #38: Mark Kistler “Imagination Station”
• #40: Distortion and deception In this case, the area covered by “Taxes” and its 3rd dimension is 1.58 times bigger than “Crude Costs” The two coloured components – the red and the green – are almost exactly the same area Image: http://www.pumptalk.ca/2008/05/sneak-peak---ne.html Discussion: https://ece.uwaterloo.ca/~dwharder/Presentations/Guidelines/VisualAids/ChartsGraphs/PieCharts/
• #41: Image: http://www.apple.com/ca/apps/iwork/keynote/ Apple Keynote example Pie 3D Wood grain (what does wood grain have to do with the categories? And what are we trying to emphasize?) This from a company that, in all other things (products and marketing) uses minimalist principles. My point: even the most rigorously devoted can succumb to this. Further discussion: http://www.presentationzen.com/presentationzen/2006/01/2d_or_not_2d_th.html http://www.theguardian.com/technology/blog/2008/jan/21/liesdamnliesandstevejobs
• #42: SOURCE: Bloomberg Business Week http://www.businessweek.com/articles/2013-04-05/who-works-and-who-doesnt-the-labor-force-by-the-numbers#r=hp-ls What’s wrong with this? Design Apparently all men, except for stay at home parents Part-time work = knitting? Colour scheme: grey scale for LF, colour for NitLF Inconsistent capitalization in captions Data representation Text caption adds to 102% (notes acknowledge rounding) but there are 101 icons (21 in top row, 20 in rest) There are 5% disabled but only 4 icons Function Hard to compare groups, eg unemployment is equal to disabled and one-third the size of retired Retired is greater than employed part time
• #43: Irony: The expression of one's meaning by using language that normally signifies the opposite, typically for humorous or emphatic effect. But we don’t want to interact with it, play with, or interrogate it in this way. Source: http://wtfviz.net/post/60373947977/you-know-youre-in-trouble-when-your-design-firm