4 of 4 people found this helpful
IMHO...The (general) purpose of a data visualisation is to allow fast (and efficient) answering of questions. Now this can be a fairly static display which is used to "show" people (in a simple to understand way) an insight you've come to, or something more interactive to allow the user/analyst to "explore" the data and answer their own questions. As such it's about using the "most appropriate" chart-type (visualisation) to answer the questions they are likely to be raised (or a specific question you want to ask of the data), and not about using the "most exotic" chart type. I have visualisations which just use simple Bar and Line Charts, as Scale and Trend (I work mainly in Retail Analytics) are generally the most important thing people ask of the data. However, I have other visualisations which use a mix of maps, and box and whiskers, where the geography and "variance" of the data is what the end-user is interested in.
My general approach to a consulting project (i.e. new client, new set of data...explore and tell them some interesting things) is to create a series of hypothesis (the questions), and then build the "appropriate" visualisation to confirm/reject those hypotheses. In practice, this is an iterative process where one hypothesis leads to another...etc. Let me give you an example. We were approached by an Airline to look at the things they sell on board (food, cigarettes, perfume...etc.). We had "initial stock" (by item), and then (at the end of the day) the "final stock" (by item), and then had a transactional table showing what had been sold during the flight. So "initial stock" - "sold stock" should equal "final stock"...but it didn't! There was some level of "unknown stock loss". So my first hypothesis was that it was being stolen (by the staff). It could well have been due to "poor transaction logging", but if that was the case I'd expect to see similar "unknown stock loss" across all products. So I build a simple bar chart showing the "unknown stock loss" by category. I found that Perfume and Cigarettes had a much higher loss than the other categories (as these are higher value, more-steelable items this added evidence to the "stolen by staff" hypothesis). I then wanted to see if this varied by airport, it did! (with different perfume brands going missing, depending which airport we looked at...French brands in French Airports, Italian in Italian...etc.). As I was looking at averages I also wanted to see the variance of stock loss (was this just one bad day, making an airport look really bad), so built some box-and-whiskers...etc. and so on. Each time it is the question/hypothesis that I think of first, and then the "best" visualisation that allows me to a) quickly answer my question b) easily show the client, why I have come to my conclusions.
For a self-explore project (building a tool for others to explore the data), it's a similar approach, where I think of the "likely" business questions the users may have, and build interactive dashboards to answer those questions (this might mean creating several dashboards specific to each type of question...which I find is much better than trying to cram the "all things to all people" mega-viz on one page)
Sorry this doesn't directly answer you question, but you need to define what is "important" to the end-user, and what questions are they likely to ask the data (eg. Does the report-medium affect the number of readers? Looking at total, average and variance/outliers), and then you can determine the best Viz to answer that question.
Hope that helps.
Thanks Simon, it helps.
I have created quite few dashboards, say 75+ with few hypothesis, slight modifications on common dashboards and as I said earlier, few charts.
And Few concepts like you said and as I said above,
My role is kinda generating ideas and concepts. Lets open this question to get more help from other experts too.
I follow Few Tableau Zen masters on Tableau Public, their dashboards are my inspirations too
Still, I want more mathematical concepts, which may or may not match with my data. But its good to know.
If you know any site, where I can learn few basic Statistics concepts which could be possible in Tableau. It would be awesome
2 of 2 people found this helpful
I would say for most of the advanced statistics, it seems people tend to use R (especially if you need muti-variate models)...as it gives them access to many statistical functions/analysis, however, there are quite a few examples out there...Here are a few I've found useful
and then Bora using some of these techniques (on Z-Score and Linear Regression) to create a very cool viz
Might not be suitable for what you need, but a really interesting article
I would also suggest using the new Tableau 10 clustering feature...could find some interesting clusters, and this (I find) leads to more Hypothesis to test.
I'd also look at anything by Bora Beran! Here's his blog Bora Beran « Worst blog article you ever saw? Well, my next one will be better.
and his fantastic article he wrote explaining how the clustering algorithm works (and why they chose the method they did)
3 of 3 people found this helpful
I agree wholeheartedly with what Simon Runc has to say, however it sounds to me like you're trying to build up a library of visualizations for future adaptations?
One of the challenges we have, as idea-generators, is we may not have a solid grasp on the visualizations you have. You may find tweaks to an existing viz can make it much more robust and demonstrative of the level of detail (no pun intended) to which Tableau is capable.
Example; Are you utilizing innate Tableau functions like the url action? You could send users to Amazon or Wikipedia to look up details on books or authors (provided these are big-named ones).
What about showing a time-graph for authors showing time between publications? (or is this Trend on Time?)
You could limit this to the top X authors and make a "race for publications" (inspired by the recent viz for Babe Ruth and the Race for Most Home Runs)
You say you have a Histogram-- but did you use the Tableau version, or a modified version which allows you to add reference lines and control bin size on the fly?
What about goofy graphs, like the relationship between length of title and number of readers? (again, this might just be a play on other types of graphs you've already made).
And no ability to do a Sankey, eh?
2 of 2 people found this helpful
Great advice from Michael Hesser ...and those "goofy" correlations/patterns often lead to some of the most interesting hypotheses (again using the clustering function can help "narrow" down which ones are "more" likely to have something interesting about them)...like the idea of the "Jump Plot" between publications, and height of "jumps" on readers! (Time to Get Hopping with Jump Plot by Chris DeMartini and Tom VanBuskirk – DataBlick (but jump plots, as far as I have seen, would require re-shaping of the data, well duplication/union of it)
Also remembered seeing this the other day
Sankey without duplication of data.
Thanks Simon. It helps alot
Few links did not go, But no problem, I can google with the title. Thanks for that
R is not in my client Machine and for now, I cannot install it.
Yes, Bora I have read few articles of his. I couldnt open his site since its a Wordpress and its blocked in my client machine I sometimes read it in my personal machine and try it when I come back work.
And Tableau 10 is not upgraded here it takes time . But I have tried simpler charts using Clusters on my personal machine. I would give a try again too.
Thanks again I feel I opened a good topic here and many of you are replying it Happy for that THanks everyone Thanks to Simon and his mentors
Yeah, my work is something like that but also few best ones will be published then and there
i have tried most of the above ones but few are new I would like to give a try with it
Bora, Andy Kriebel, Tableau Public authors - Pooja Gandhi, Many are my inspirations (I might even be following you )
The above are good collection for me to work for another few days If you know someone I can follow, Please share their Tableau Public Profile link/Site too
Sankey without editing the data source, thats great Lemme try that
Jump plots looks great,but editing data source, not possible will try it and see if there is any learnings from it for me
2 of 2 people found this helpful
One more to add to your Tableau Public follow list...in case you haven't come across his work (Nicco Cirone)
Very imaginative and very beautiful Visualisations