Hi! I'm Andy Cotgreave: co-author of The Big Book of Dashboards, and Senior Data Evangelist for Tableau, a Salesforce company. I am passionate about how we communicate with data.
I co-host Chart Chat, the monthly video series about data communication, and write the Sweet Spot newsletter.
I've got over 17 years’ experience in the industry, and have helped thousands of people improve their data cultures. Subscribe for content that entertains, informs and educates.
The other issue is that the Swingometer was designed in an era when most people voted Labour or Conservative. Even showing swing for an individual seat was unhelpful since there was not necessarily an increased vote for an increased or reversed margin.
Hi Andy , great work and I liked the way you compared and explained. If the dataset was over a Gigabyte, was it possible to upload it to GPT and the analysis it? Thanks
Now that ChatGPT 4 has access to Python, its math capabilities are much better. It now gets the math correct. As for identifying the date where B begins consistently outperforming A, you need to clarify the prompt. Here's what ChatGPT suggested: To avoid similar mistakes in the future, it's essential to verify and cross-reference the data carefully, especially when making claims about trends or consistency. A recommended prompt to help with accuracy might be: "Please provide a month-by-month comparison to determine from which month Product B started to have higher sales than Product A without any interruptions in the trend." This prompt encourages a detailed and careful analysis, prompting the assistant to look at each month in sequence to verify the continuity of the trend. Additionally, it might be helpful to ask for a summarized view of the comparison, such as: "Can you confirm if there were any months where Product A outsold Product B after the trend for higher sales for Product B started?" This follow-up can serve as a double-check to ensure the trend identified is accurate.
dynamic commenting- Ok, so finally got the presentable charts. @39.29 circle is the most commonly reported shape, the last sentence on the screen reads it wrong. Also, the final presentation mentions it wrong - reporting 'light' as the most common shape.
sure it looks cool but i think the audience for this will be small. unless you're prepping for a TED-style talk this would be way too much time and effort to prep for your average analyst.
I don't disagree! My optimism though comes from the number of people who create powerpoints, but want to do better when showing data. For those people, they'll set themselves apart.
Can you tell me that coming in data analytics line after coming chatgpt4 is good or not??? I'm from non technical background having no degree in computer science
Thanks for sharing Andy, this is so interesting! Wondering how many folks using ChatGPT to do their work/schoolwork that would be so sad when they see this!
On the subject of ChatGPT augmenting what a person does, as you mention at the end ... true enough, though one of the things the person would be most likely to rely on the machine doing is the arithmetic part, which it got wrong. This takes "How to Lie With Statistics" to a new level. :-)
I tried to watch with an open mind, but I suspected it would not be able to do actual analysis. I am very surprised it didn't do arithmetic correctly, though! I was also surprised that a line chart was not the first suggestion. Though I would choose another chart type for this data, I would have started my EDA with a table and a line chart. Ultimately I would've chosen something highlighting the differences between the two products by month, like you did with the dumbbell chart. My favorite option the human came up with was the variance from a zero baseline. It's the most informative if I want to see the difference between products AND changes over time.
When I do that exercise in sessions the majority choose a line a chart first so I was surprised to see that as its third choice. Side-by-side bars aren't great.
@@AndyCotgreave I broke down and decided to try the damn thing this morning. I'm doing the biggest dataviz project I have ever done and I've personally collected a lot of the data, so I'm going back to fact-check myself on a few things. I asked ChatGPT who the murderer was in an Agatha Christie novel. It's available online as full text in multiple places, so I was thinking, "here's a softball, ChatGPT." It got it wrong the first time, identifying a murderer from one of her other novels. When I broke the news to it, it apologized and changed it's answer to a character from a different book. On a third try, it gave me a character from the correct book, but not the murderer. I could've Googled it so much faster, but I decided to give it a go.
ChatGPT says "it depends"! 😍 In experimenting with this in recent weeks I've been struck by just how explicit a user needs to be with their commands/questions. I wonder, in this case, whether it has interpreted the month names as a category primarily, rather than dates within a single year (as we would naturally interpret them). I'd be curious to explore what it might do if we were more explicit as to the properties of the 'Month of Data' field. Or refer to those dates as January 2022, February 2022 etc, so that it understands that it can interpret them as continuous dates with more confidence.
I had an attempt at reproducing this with the dates in MMMMM YYYY format, and got this response, which hints that we can provide it with better guidance as to what the data type is and get a better result out: "Based on the dataset provided, a line chart would be the best chart to use to visualize the sales trends for each product over time. The line chart would have months on the x-axis and sales on the y-axis, with separate lines for each product (A and B). This would allow viewers to easily see how each product's sales have changed over the year, and to compare the two products' sales trends side-by-side. Additionally, you could consider adding markers for each data point (i.e., each month) to make it easier to see the exact sales figures for each month. Another option to enhance the visual appeal of the chart would be to add color to differentiate the lines for each product, for example, using blue for Product A and orange for Product B."
the swingometer is an essential part of election night - dropping it would be ludicrous, better use however needs to be exercised (i.e. not using it on every seat and it should absolutely be to scale) all very valid criticisms highlighted here but swingometer are a key part of election night
additionally, vine is useless on the swingometer, and the design has been extremely poor since 2005 - focused more on trendy graphics than displaying the data!
The thing that always amazes me is the fact that despite how horrendous the conditions were during the retreat, to the point that only a few % made it back alive, these survivors STILL fought on bravely and skilfully at Leipzig not long after... im surprised they even had the strength to walk, let alone fight after what they endured
you know that if you tried to escape you would be killed right ? What amazes me is that a man who sends 410 thousands young men to murder, rape other people, and then die, is still seen as a hero today. That scares me for the future.
I don't think people think he's a hero, more of a curio and a phenomenon, isn't that it, that no single person has ever had more might at his disposal, and used it the most insanely ? Weininger considered him the zenith of criminality. And said he was also an epileptic, something never mentioned. Anyone know?
