The topic for week #4 was New Zealand tourism spending. Tableau Public Link for this Submission
Square Tile Map
Originally, I intended to use the extracted shapefile data provided by #MakeoverMonday. However, an error was occurring for the southernmost district, which inverted the shape in a very small area. After several attempts to create a Band-Aid solution, this effort was abandoned.
Instead, I decided to abstract the geospatial information by using a tile map, which requires prep work in Excel to create the x/y coordinates. First, I rounded latitude to break up the records into about a dozen rows on the y-axis. Next, an x-value was assigned by ranking longitude. This was a manual process, which resulted in a new Excel table. Once completed, an inner-join was created between the original data source and this new tab.
I believe most people using a grid style went with a big map and a line chart within each square. My preference is to go small and make more of a heat map effect, because I think it’s easier to see the trends this way. Another example of this heat grid map approach can be seen in my #MakeoverMonday submission for bad driving trends.
The color rules were another choice that impacted the map design. Instead of using a continuous variable, which limits design control, a discrete value grouped items based on standard thresholds. The hope was this small technique would make the patterns easier to see.
Additional Data Prep
Not being a fan of total RTI being caught up in the data pivot, I decided to create a new table with this value joined back into the original data set. Pivoted data should be an apple to apples comparison, and total RTI in the original data set did not pass this rule. This small data prep step allowed me to create the following calculation:
IF COUNTD([REGION]) > 1 THEN [Total RTI] ELSE [Regional Tourism Indicator (baseline 100)] END
This simple condition ensured total RTI would display when no selections were made, while avoiding any additional aggregation from Tableau.
Although I abandoned the geospatial file, I decided to keep the population and area data points. Population density categorized the cities and districts. This was a major design decision that clearly separated my approach from a more RTI-driven navigation.
My reasons for not showing a full distribution of all available districts with their RTI in one visualization:
- If building for expected utility, then the majority of people would likely want to see the major population areas first. So, the navigation defaulted to what I expected to be the popular selections. I wanted it easy for users to find the major population areas, like Auckland and Christchurch.
- Giving equal visual representation of Auckland with a sparsely-populated rural district could potentially undervalue the importance of Auckland’s impact to overall RTI when analyzing visually.
- The total RTI metric seemed to work great for summarizing the data at a higher-level.
I admit the UX was complex in places on this dashboard, but my main goal was to prevent additional aggregation. I worked on several projects with aggregated survey data, so this is a process that I am familiar with. Every chart shows data that matches its underlying data, and the logic simply locates the desired records from different user selections. Analysts, who created this aggregated data set, already completed the difficult calculations, so let’s not try to make it more complicated. 🙂
This limited approach to dashboard design minimizes the chance for error in a number of areas. For example, even if I included the partial 2016 data, my line chart and tile map would have handled this inclusion. Unfortunately, the candlestick chart required an equal distribution, so 2016 was left out.
Undeniably, a peculiar choice. The selection was partly influenced from my #MakeroverMonday work for week #3 and partly driven by the data set. Since we are working with highly seasonal aggregated data, I knew several restrictions existed for visualizing annual trends. The twelve data points for each year create the annual min and max range, while the month parameter creates the middle box. The color communicates the direction of change, and the box’s size indicates the magnitude of change compared to the selected month’s prior year. This min/max range shows a broader trend, but more importantly, this range provides a scale for each monthly selection. By displaying seasonality in a separate line chart, my attention was free to focus on the annual trends and change with the candlestick chart.
Making the candlestick chart parameter-driven made it a flexible way to visualize complex time-series data. Essentially, this implementation of a candlestick chart was a hybrid between the box plot, which communicates distribution, and the traditional candlestick chart, which communicates change and trend. If designed effectively, then I thought this had the potential to provide a lot of information to my audience. Relying on people to interpret an unfamiliar chart type is always a dangerous game, but I decided it was worth the risk. Every familiar chart had been seen for the first time at some point in history. 🙂
My tinkering with the candlestick chart motivated me to tinker anew with the line chart. I went through many iterations of different line charts, and I was unhappy with all them. One of the more interesting failures is displayed below:
Since we could not average the indexes, I considered three main approaches. First, use a year parameter to select a single year at a time. This is a sound decision from a data visualization perspective, but one would not be able to make comparisons with other years. For the second option, show all months and years in a line chart to allow comparisons. This resulted a less than useful line chart due the large seasonality. It didn’t communicate seasonality or the overall trend particularly well. The third choice was to take the best from both of these approaches and avoid the drawbacks. The selected year is clearly visible, while non-selected years are in the background to provide context.
Originally, I used line charts for both the selected and non-selected years. So, the size marks and transparency rules were solely responsible for encouraging focus in the appropriate area. I fully expected this to make it through the final edits, but I eventually scuttled this version. By random thought, I tried using circle for non-selected years for the background chart instead of lines. This was probably my most difficult decision, and I lack conviction or confidence between these two approaches. The hope was for the circle chart to appear more distinct from the line selection, but looking back, it’s not clear how important this consideration should have been.
The size marks used three values, which is probably unusual. The small-sized mark equaled 1, the medium-sized mark equaled 2, and large-sized equaled 4. The transparent and smallest marks indicate the data records are not associated with the years selected. The middle-sized marks match the year, but do not match the month parameter. The large-sized mark indicated the single record which matched both the month and year selection. This helped support a flexible and interactive chart.
A significant danger existed for overwhelming users with so much data in a compact space. So, similar blue colors and transparency were used as much as possible to soften the overall look. Likewise, all reference lines were as soft as possible, while still performing their function.
The color scheme used in the first version included grey, black, and red, because I originally wanted to go with a lighter background. However, I decided to switch to a dark blue background for three major reasons. First, grey is my favorite color, and I seem to rely on it too much. Second, the dark blue color gave it a formal feeling, which seemed to be a better fit for economic data. Third, dark blue would give me a lot of flexibility in managing a user’s attention with light blue for subtle contrast and gold/orange for starker contrast. The tile map also looked best with this blue color scheme.
This year, I have a simple rule regarding the use of graphics in my dashboards. If I add any graphic to my data visualization work, then I have to make the graphic myself. This rule forces me to be more purposeful in image choices, helps make my work more unique, and gives me an additional outlet to be creative.
Week #4 was not an exception to this rule. Designing New Zealand currency seemed like a good choice, because it relates to economic data and contains a number of national symbols. The gold and silver colors worked well with the dark blue background, but this was not due to my intention. Additionally, the top three stars from the Southern Cross constellation finished out my chart art, which also worked well with the midnight blue color.
My Tableau experience still far outpaces my Illustrator experience, so I am trying to practice more with both programs. I fully expected these coins to not turn out well, because it seemed like an ambitious project for my current skill level. The whole design process for the Illustrator work this week is worthy of its own blog post. Below are the final designs before I uploaded each one to Tableau.
Every #MakeoverMonday project is an interesting and challenging experience for me. The number of decisions made throughout the creative process is astounding and impossible to capture in a single blog post. I am always looking forward to the next project and to continue on my Tableau journey.